Skip to content

Methodology · Wi-Fi gateways and access points

TR-398: what a gateway delivers in a real home, measured in a shielded box with nothing else on the channel

Broadband Forum TR-398 is the test plan ISPs and vendors use to accept a Wi-Fi gateway, access point or mesh node. Five groups of tests: RF capability, baseline performance, coverage, multiple stations and long-term stability, each with a minimum for the device class. We run it on request on our Wi-Fi bench, with emulated stations up to Wi-Fi 6 behind programmable attenuators in a shielded enclosure; the full chamber set-up is agreed per project. Wi-Fi 7 is in progress on the bench and not reported yet.

  • Every test with its own pass line, per band and per channel width, from sensitivity to a four-hour run
  • Spatial consistency and AP coexistence are agreed per project, with the setup named in the report
  • Pass thresholds from the TR-398 tables for the device class, your own figures on top where the tender names them
Get a TR-398 test plan Jump to the test groups
RANGE VERSUS RATE 5 GHz · 2×2 · 80 MHz · DOWNLINK MEASURED PER STEP TR-398 MINIMUM BELOW THE LINE 1,120 1,010 840 580 300 92 Mbit/s 10 dB 20 dB 30 dB 40 dB 50 dB 60 dB Throughput at each attenuation step, each against its own minimum

Range versus rate for one band and one channel width: throughput at every attenuation step against its own minimum; one step below the line fails the test

Where it fits

Acceptance of the Wi-Fi device, not of the Wi-Fi standard

RFC 2544 proves what the wired ports forward. TR-398 proves what a household gets from the radio: at range, with many clients, with an old laptop in the room, for hours

ISP gateway acceptance

The box that goes into ten thousand homes with the next firmware. The report says which test groups passed on which band, with the thresholds the tender named, before the rollout

Vendor bake-off

Two or three candidate gateways or access points measured the same way, in the same enclosure, with the same stations. One table, one winner per criterion

Firmware regression

A new driver, a new channel plan, a new mesh release. The baseline set rerun against the previous report shows what moved, before the field finds it

Test bed

Stations on one side, the wired network on the other, the device in a shielded box between them

TR-398 asks for a controlled RF path. Our bench feeds the stations through programmable attenuators and keeps the wired side at line rate, so the radio is the only variable

STATION EMULATOR · WIRELESS SIDE EMULATED STAs · 802.11n / ac / ax WHAT EACH STATION BRINGS Its own radio, streams and channel width Rate, PER, RSSI and retries per STA A legacy 802.11n client when the test asks UDP and TCP, up and down, per station UP TO WI-FI 6 TODAY · WI-FI 7 IN PROGRESS RF CABLES ATTENUATORS DUT IN A SHIELDED ENCLOSURE GATEWAY · AP · MESH NODE, AS SHIPPED 2.4 GHz · 20 / 40 MHz11n / 11ax 5 GHz · 80 / 160 MHz11ac / 11ax PATH LOSS PER STEP10 to 60+ dB NO ROOM, NO NEIGHBOURS IN THE NUMBERS PATH LOSS CALIBRATED TURNTABLE ON REQUEST SECOND AP ON REQUEST DUT LOGS, WHOLE RUN LAN / WAN ≤ 10 GbE WIRED GENERATOR · WIRED SIDE THE OTHER END OF EVERY FLOW WHAT THE WIRED SIDE GIVES Line rate at the LAN port, never the limit Throughput and loss per station, per second Latency and jitter across the air link Same clock as the station emulator THROUGHPUT · PER · FAIRNESS · TIME BED FIRST, DEVICE SECOND path loss of every cable and attenuator measured with a reference station before the DUT goes in; the offsets are in the report

Station emulator

Emulated stations with their own radios, streams and channel widths, 802.11n, ac and ax up to Wi-Fi 6. Rate, PER, RSSI and retries per station every second; a legacy station added to the mix when the test calls for one

Device under test

Gateway, access point or mesh node in a shielded enclosure, configured as shipped, both bands on. Path loss of every cable and attenuator is calibrated with a reference station before the device goes in, and the offsets are in the report

Wired side

A traffic generator on the LAN or WAN port at 1 or 10 GbE, the other end of every flow. Throughput and loss per station, latency across the air link, on the same clock as the station emulator

What the bench does today

Wi-Fi 6 on 2.4 and 5 GHz. Wi-Fi 7 is in progress and is not reported until it is. A turntable for spatial consistency and the full chamber procedure are agreed per project; the report states which setup was used

Parameters

What is fixed before the run

TR-398 fixes the test setup and the pass thresholds per device class. The plan lists what we set on top, with the value we use when the tender is silent

Parameter Our default Allowed Why it matters
Device classFrom the datasheet: band, streams, width2.4 GHz, 5 GHz, 2×2 to 4×4, 20 to 160 MHzEvery TR-398 threshold is a function of the class; a 2×2 gateway judged by a 4×4 line fails for the wrong reason
Channel and widthAs shipped, auto offAny channel of the regulatory domainA device left on auto may change channel mid-run; the plan pins the channel and the report names it
Station types802.11ax 2×2, plus 802.11n for fairness802.11n, ac, ax; 1×1 to 4×4The stations decide what the device can show; a legacy station is the whole point of the airtime fairness test
Attenuation steps10 to 60 dB in 10 dB stepsAny list inside the attenuator rangeRange versus rate is a curve, and the minimum applies at every point of it; finer steps show where the rate collapses
TrafficUDP for rate, TCP on requestUDP, TCP; down, up, bothUDP shows the radio; TCP shows what a download gets after retries and window growth, usually lower
Station count32 for the multi-STA test1 to the bench limitA home has twenty to forty clients; the sum rate and the per-station floor both have to hold
Long-term run4 h4 h or longerResets, memory leaks and thermal throttling show after hours at load, not in a 60 s trial
Pass thresholdsTR-398 table for the classTable values, or stricter ones from the tenderEvery test is pass or fail against its own line; one miss marks the group failed and the run goes on

Thresholds follow the TR-398 tables for the device class. When the tender names a stricter figure, the tender wins; a looser one is reported next to the table value

Always in our runs

  • Path loss calibrated before the DUT goes in
  • Both bands, DUT configured as shipped
  • Thresholds from the TR-398 table for the device class
  • Per-station counters every second, both sides
  • Wired side at line rate, never the bottleneck
  • Console log and DUT logs for the whole run

Calibrated path loss, both bands, the device as shipped, thresholds from the table, counters per station from both sides, console and logs for the whole run

Agreed per device

  • Baseline set only, or all five groups
  • Turntable for spatial consistency
  • Wi-Fi 6 features on: OFDMA and MU-MIMO
  • Client mix with a legacy 802.11n station
  • Long-term run: 4 h, or longer
  • Second AP and its channel for coexistence

The set of groups, the turntable, Wi-Fi 6 features, the client mix, the length of the long run and the coexistence setup are fixed in the plan with you. Each one changes the bench days and what the report can claim

Inside the method

Five groups, from the radio to a four-hour run

TR-398 builds up: first the radio alone, then throughput, then range and angles, then many stations, then time and a neighbour on the channel

  1. 01RF capability

    Receiver sensitivity and maximum connection

    The station level is lowered step by step until the packet error rate crosses the limit; the level where it still holds is the sensitivity, per band and per rate. Maximum connection checks that the device accepts the number of stations the class requires and keeps them

    Per band · PER at each level
  2. 02Baseline

    Maximum throughput, dual-band, bidirectional

    One station at short range, down then up, then both at once, then both bands loaded together. Each figure against the table minimum for the band, the stream count and the channel width

    Per band · UDP, 2 min per run
  3. 03Baseline

    Airtime fairness

    Wi-Fi 6 stations at full rate, then a slow 802.11n or 802.11ac station joins at range. A fair device gives the slow client its share of time, not its share of frames; the fast stations must keep the rate the table names

    5 GHz · legacy client in the mix
  4. 04Coverage

    Range versus rate, spatial consistency

    Attenuation rises in steps and the rate at each step must stay above its line. On a turntable the device is turned through the angles and the rate must not fall below the limit in any direction; an antenna that favours one wall shows here

    10 to 60 dB · turntable when agreed
  5. 05Multiple STAs

    Multiple stations, join and leave, MU-MIMO

    Many stations at once, 32 by default, with a sum rate and a per-station floor to meet. Stations associate and disassociate in a loop while the others carry traffic; no station may be lost. Downlink MU-MIMO to several stations at once when the device claims it

    32 STAs · both bands
  6. 06Stability

    Long-term stability, AP coexistence

    Four hours at load with the rate sampled every second: no reset, no reassociation, no throttling beyond the tolerance. Then a second access point on the same channel with its own traffic; the device must keep the rate the table names for a shared channel

    4 h · second AP on the channel

Procedure

From the plan to the signature

The baseline set fits in a bench day; coverage, stations and the long run take one or two more. The whole plan for one gateway is two to three bench days

  1. 00

    Test plan

    Device class, bands, channels, station types, the groups to run, the length of the long run and every threshold in one table, signed off before the device goes into the enclosure

  2. 01

    Bed calibration

    Path loss of every cable and attenuator measured with a reference station on both bands. The offsets go into the plan; the device goes in only after the bed is known

  3. 02

    Baseline set

    Sensitivity and maximum connection, then maximum throughput down, up and both ways, dual-band, airtime fairness. A failed test is reported the same day so the firmware can be changed before the rest runs

  4. 03

    Coverage, stations, time

    Range versus rate in steps, angles on the turntable when agreed, thirty-two stations, join-and-leave loops, MU-MIMO where claimed, then four hours at load and the coexistence run

  5. 04

    Report and hand-over

    Pass or fail per test and per band, with the rate at every step, the per-station series and the DUT logs attached. A failed test comes with the level, the station mix and the minute

ONE DEVICE · FIVE GROUPS · TR-398 BASELINE FIRST, THEN RANGE, STATIONS AND TIME · EACH TEST AGAINST ITS OWN LINE BASELINE SET · DAY 1 RF CAPABILITY + BASELINE · 2 BANDS SENSITIVITY THROUGHPUT FAIRNESS level down to the PER limit, per band down · up · both · 2.4 + 5 GHz legacy client joins COVERAGE · STATIONS · STABILITY DAYS 2 AND 3 RANGE STAs 4 h+ attenuation steps · angles on the turntable 32 stations · join and leave · MU-MIMO long run at load · second AP on the channel RATE ≥ TABLE MINIMUM AT EVERY STEP = PASS SLOW CLIENT GETS ITS SHARE, NOT THE AIR = FAIR A RESET OR A LOST STATION IN THE LONG RUN = FAIL

One device, five groups: the baseline set first, then range, stations and time. Every test has its own line from the TR-398 table, and the run goes on after a miss

Pass rules we apply

  • PER ≤ limit

    At each sensitivity level of the table the packet error rate stays within the limit; the first level that breaks it ends the test for that rate

  • ≥ table rate

    Throughput at every step of every test meets the minimum for the device class, band and width; a stricter figure from the tender replaces the table

  • Fair share

    With a legacy station in the mix the Wi-Fi 6 stations keep the rate the table names; a device that lets the slow client own the air fails here

  • 0 drops

    No reset, no lost station and no rate outside the tolerance in the four-hour run; a single reassociation is logged and fails the group

Time budget

  • Bed calibration2 hours, once per set of cables, attenuators and bands
  • Baseline setOne bench day for both bands, including a rerun after a firmware change
  • Coverage and stationsOne bench day; add half a day for the turntable
  • Stability and report4 hours at load plus the coexistence run; report the next business day

Report

What you get at the end

A verdict per test and per band, and the series behind it. Enough to accept the firmware, pick the vendor, or send the developer to the exact level and station mix

Per test

  • Sensitivity level per band and rate, with the PER at each step
  • Throughput down, up and both ways, per band, against the table minimum
  • Rate at every attenuation step and, with the turntable, at every angle
  • Per-station rate and the sum in the multi-STA run, join-and-leave events
  • Four-hour rate series with resets, reassociations and throttling marked
  • Pass or fail against each threshold, table value or tender value

Context

  • Device class, firmware, channel, width and security mode as tested
  • Bed calibration: path loss offsets per cable and attenuator, per band
  • Station emulator model, station types and the legacy client used
  • Which TR-398 groups were run, which were left out and why
  • Enclosure or chamber, turntable yes or no, second AP and its channel
  • Timestamps of every run and the DUT log around every failure
TR-398 · REPORT GATEWAY · 2×2 · WI-FI 6 01 Receiver sensitivity · 2.4 and 5 GHz pass PER within the limit at every level of the table 02 Maximum throughput · 5 GHz down 1,118 M 80 MHz · UDP · table minimum for the class met 03 Airtime fairness · one 11n client −62 % Wi-Fi 6 stations fall to 38 % of their alone rate 04 Dual-band and bidirectional pass both bands loaded · up and down at once 05 Range versus rate · 6 steps 6 / 6 rate above the minimum from 10 to 60 dB 06 Multiple STAs · 32 stations pass sum rate and per-station floor met 07 Long-term stability · 4 h 0 drops no reset, no reassociation 08 AP coexistence pass second AP on the channel · rate within the limit ATTACHED RATE PER STEP PER-STA CSV 4 h SERIES DUT LOGS · PCAP Values are illustrative. Your report carries your device class, thresholds and groups.

Attached

  • Rate per step and per station as CSV, for your own charts
  • Four-hour time series, DUT console log and a PCAP around each failure
  • Range and fairness graphs as PNG, ready for the acceptance document

From practice

The gateway that passed on speed and failed on manners

A Wi-Fi 6 gateway for an ISP fleet, 2×2 on 5 GHz. Maximum throughput passed on both bands with room to spare, range versus rate cleared every step. Then the multi-station run with one 802.11n laptop added at range: every Wi-Fi 6 station lost 62 % of its rate, well below the airtime fairness line.

The slow client was sending at a low rate and taking most of the air for it. Airtime fairness existed in the firmware and was off by default. One setting, one rerun: the Wi-Fi 6 stations came back to 94 % of their alone rate, the old laptop kept its share. The fleet shipped with the setting on.

TR-398 airtime fairness and multi-STA runs on a 2×2 Wi-Fi 6 gateway, 5 GHz, 80 MHz

SAME GATEWAY · RATE PER WI-FI 6 STATION WI-FI 6 STATIONS ALONE 5 GHz, 80 MHz, four 2×2 stations · maximum throughput passed 268 M ONE 802.11n STATION JOINS the slow client holds the air · each fast station loses 62 % 102 M AIRTIME FAIRNESS ON IN FIRMWARE · RERUN 251 M One setting, off by default. Found on the bench, not in the field

Services

How we run TR-398 for you

On our Wi-Fi bench with your device shipped to the rack, the baseline set or all five groups. Same plan, same report

Gateway acceptance for an ISP

The groups the tender names, on the firmware that will ship, with the thresholds written into the contract. The report is the acceptance document

Order an acceptance run

Bake-off of candidate devices

Two or three devices through the same plan in the same enclosure, one table at the end and a winner per criterion the buyer set

Set up a bake-off

Firmware regression

The baseline set rerun on each release against the previous report. What moved, by how much, on which band

Bring the next build

Bench access for your team

Your engineers run the plan on our bench through CloudLAB, with our engineer on chat and the device in our enclosure

Open CloudLAB
  • Report in the TR-398 structure, group by group, band by band
  • Station emulator and attenuators chosen for the device class, from more than one vendor
  • Engineers who read the tender thresholds before the counters

Related pages

FAQ

Questions we get about TR-398

Is your lab a certified TR-398 test house?

No. We run the TR-398 procedure on our own Wi-Fi bench: emulated stations up to Wi-Fi 6, programmable attenuators, a shielded enclosure, and the rest of the chamber procedure agreed per project. The report states exactly which setup was used and which groups were run, so a buyer can see how close it sits to the full chamber procedure. For a certificate from an accredited house we help write the plan and read the result.

Do you test Wi-Fi 7 devices?

The bench supports stations up to Wi-Fi 6 today; Wi-Fi 7 is in progress and we do not report Wi-Fi 7 figures until it is. A Wi-Fi 7 gateway can be tested now on its 802.11ax modes, with the limitation written on the first page of the report.

How long does a full TR-398 run take?

Two to three bench days for one device: a day for the baseline set on both bands, a day for coverage and the multi-station tests, four hours or more for the long-term run plus coexistence. A bake-off of three devices takes about a week. The baseline set alone fits in a day.

Which devices does it apply to?

Residential gateways, ISP-supplied access points, mesh nodes and enterprise access points used in home offices. TR-398 was written for the device the operator hands to the subscriber; for an enterprise campus the same bench runs a client-mix plan without the TR-398 thresholds.

Can we test mesh or multi-AP systems?

Yes, on request and as a separate plan. The basic set tests one device in the enclosure. For a mesh we put the root and one or two satellites on calibrated paths and measure the rate at the far node, roaming between nodes and the backhaul share; the thresholds come from your requirements, not from the TR-398 table.

What happens when the device fails a test?

The run goes on and the test is marked failed with the level, the station mix and the minute. You change the firmware or the configuration, we rerun that test or the whole group, and the report shows both results side by side. Reruns on the same setup are bench time, not a new project.

Test plan and quote in 1 to 3 business days

Send the device model, the firmware and the groups the tender names. You get the plan with every threshold, the bench days and a quote for the run

Request a TR-398 plan Log in to CloudLAB

Send the shortlist

You have the method. Now get the numbers.

Name the models and the KPIs that matter. An engineer replies within one business day with a test plan draft and a quote.

Prefer to chat? Message an engineer on WhatsApp

We use what you send only to answer you. See our Privacy Policy.