Reading a Site Survey

A site survey produces a picture with colors on it, and the picture is the least important part. What matters is whether the numbers behind the colors meet a stated target at the places people actually sit.

Most districts skip the survey, deploy APs where the ceiling tiles were convenient, and then spend years tuning around a design that was never measured. This article is about measuring it once, properly.

Three kinds of survey

Type	What it does	When
Predictive	Models coverage from a floor plan and wall materials	Before construction or cabling
Passive	Listens to existing APs; shows signal and interference	Auditing an existing network
Active	Associates and measures throughput, roaming, retries	Validating a design, or diagnosing

Predictive surveys are cheap and wrong in useful ways — they tell you where to put cable drops. Passive surveys tell you what the air looks like. Active surveys are the only ones that answer "will a Chromebook work here."

Do all three in that order for a new building. For an existing one, start passive and active.

The numbers that matter

Metric	Target	Why
Primary RSSI	−67 dBm or better	Enough rate for voice and video
Secondary RSSI	−72 dBm or better from a second AP	Roaming without a drop
SNR	25 dB or better	Higher MCS rates, fewer retries
Noise floor	−92 dBm or lower	Anything above −85 is a problem
Channel overlap	Under 20% of the floor	Co-channel contention
Retries	Under 10%	The real health metric

If a heatmap shows green everywhere and retries are 25%, the survey measured the wrong thing. Coverage is necessary; it is not sufficient.

Wall materials change the plan

Material	Attenuation (approx, 5 GHz)
Drywall	3–5 dB
Brick	6–10 dB
Concrete	12–20 dB
Metal studs / mesh	15–25 dB
Elevator shaft, stairwell	Effectively a wall

Two classrooms separated by a cinderblock wall are two cells, not one. A media center with a metal shelving grid will have dead spots that a predictive survey never predicted.

Capacity is a separate design

Coverage answers "can a device connect?" Capacity answers "can 900 devices connect at once and still work?" They produce different AP counts.

A rough planning number for a classroom wing: 30–40 devices per radio at 5 GHz with 40 MHz channels, and fewer if the devices are doing video. A 1:1 elementary campus with 700 devices needs roughly 20–24 APs for capacity, not for coverage — coverage alone might be satisfied with 12.

Worked example: validating a wing

Four APs, staggered 40 MHz channels, 14 dBm. Walk the corridor and two classrooms per AP:

1. Primary RSSI ≥ −67 dBm at every student seat. If not, the AP is in the wrong place or the power is too low for the wall type.
2. Secondary RSSI ≥ −72 dBm so a device can roam before it drops.
3. SNR ≥ 25 dB; check noise, not just signal.
4. Retries under 10% while a class-sized load runs (20 devices streaming).
5. Roam test: walk the corridor with a call or a continuous ping; count drops.

If step 5 shows more than one or two lost packets, the problem is usually coverage overlap, not roaming configuration.

What breaks in production

• Surveying at full power, then deploying at low power. The heatmap describes a network you did not build.
• Surveying an empty building and treating the result as final.
• Coverage without capacity. Green everywhere, unusable at 10:00.
• Ignoring the uplink. An AP with a 1 Gbps uplink and 600 Mbps of radio throughput is fine; one on a 100 Mbps link is a bottleneck the heatmap cannot show.
• Trusting a vendor's predictive model without a validation walk after installation.

Checklist

1. Every seat has a measured RSSI and SNR, not a color.
2. Secondary coverage exists everywhere a device might roam.
3. Capacity (devices per radio) was designed separately from coverage.
4. The survey reflects the building as used: doors, furniture, people.
5. Retries and roaming were measured under load, not at idle.

Next

Coverage and capacity are designed. Continue with Roaming in Practice for what happens when a device walks from one cell to the next.