Skip to content
ARRF
Get Support

What the phone measures

On an Any Wi-Fi survey — and on a RUCKUS survey with Measure experience on — the iPhone itself times your network as you walk. It probes your router (the Wi-Fi hop you can fix), a nearby test server on the internet, and runs short speed bursts. Each cube summarises the probes from the 2 seconds before it was placed. None of this needs access to the access points, so it works on any Wi-Fi from any vendor.

These tests run on Wi-Fi only; if the phone drops off Wi-Fi, the survey pauses measuring until it’s back.

Which metrics are there, and how is each measured?

Section titled “Which metrics are there, and how is each measured?”
Metric (colour mode)How it’s measuredNeeds internet tests
Wi-Fi latency (p95)ICMP echoes to your router, 20 a second. The 95th-percentile round trip of the idle replies in the cube’s window. If the router ignores ICMP, ARRF falls back to timing TCP connects.No
JitterThe average change between consecutive idle router round trips.No
Packet lossThe share of idle router probes with no reply within 1 second.No
Internet latencyWebSocket echoes to the test server, 10 a second; the median round trip.Yes
Download ↓ / Upload ↑Short speed bursts against the test server, or a local iperf3 server if you set one.Yes, unless you use local iperf3
Latency under loadThe median router round trip for probes sent during a speed burst.For the burst
Voice-class latencyThe median round trip on the simulated-call lane (WMM voice queue).No
Call quality (MOS)A simulated call to your router, scored with the E-model — see below.No
Call quality under loadThe same call score while a speed burst is running.For the burst
Internet call quality (MOS)The same call frames sent end to end to the nearest Cloudflare edge.Yes
DNS lookupOne DNS query a second to the phone’s own resolver; the median time.No
Connection setupA fresh TCP + TLS connection to the test server, up to once a second; the median setup time.Yes
RetransmissionsFrom a speed burst: the worse of the upload’s retransmitted share and the download’s duplicate/out-of-order share.For the burst
OS link qualityiOS’s own rating of the Wi-Fi link: minimal, moderate, or good. iOS 26 or later only.No
Connected AP (BSSID)Which access point (BSSID) the phone was on — one colour per BSSID. Needs Location access.No

Idle figures never include probes sent during a burst; those become the under load figures instead.

During a continuous survey ARRF runs a 1.5-second burst about every 5 seconds, alternating download and upload; a hot phone bursts less often. In an Any Wi-Fi stop-and-sample capture, the phone measures for one window and runs a download and an upload burst before the cube drops. Bursts go to the default test server, or to a local iperf3 server set in Settings → Experience, which measures speed over your own network only, without the internet in the way. Saturate (latency under load) in the same settings keeps a transfer running instead, so every cube also measures latency under load; it uses more data and battery.

The call test sends 50 packets a second — one G.711-sized frame per 20 ms — to your router, marked for the WMM voice queue. ARRF scores it with the ITU-T E-model for G.711 with packet-loss concealment, giving a MOS (mean opinion score) from the lane’s delay, jitter and loss.

MOSLabel
4.3 or moreExcellent
4.0 – 4.3Good
3.6 – 4.0Fair
3.1 – 3.6Poor
below 3.1Bad

This score covers this iPhone to your router — the Wi-Fi hop — not an end-to-end call. Internet call quality sends the same frames to the nearest Cloudflare edge over a TCP WebSocket; any frame more than 60 ms later than the fastest ones counts as lost, as in a real call’s jitter buffer.

The Results Voice card also reports Voice priority (WMM): it compares how much the voice lane and normal traffic slowed during the speed bursts. A failure means priority isn’t kept somewhere between this phone and the router.

Each metric paints cubes on a red → green ramp with four breakpoints, read as Poor / Fair / Good / Excellent. A value exactly on a breakpoint counts toward the better side. For lower-is-better metrics the breakpoints run downward.

MetricPoorFairGoodExcellent
Wi-Fi latency (p95), Voice-class latency100 ms50 ms20 ms10 ms
Jitter20 ms10 ms5 ms2 ms
Packet loss5 %2 %1 %0.5 %
Internet latency150 ms80 ms40 ms20 ms
Download ↓, Upload ↑3 Mbps10 Mbps25 Mbps100 Mbps
Latency under load200 ms100 ms50 ms25 ms
Retransmissions10 %5 %2 %1 %
Call quality, under load, Internet call (MOS)3.13.64.04.3
DNS lookup150 ms60 ms30 ms15 ms
Connection setup400 ms200 ms100 ms50 ms

OS link quality and Connected AP are categories, not ramps: link quality runs minimal → moderate → good, and each access point gets its own colour.

The colour menus lead with three presets, then All metrics:

PresetModes
VoiceCall quality (MOS), Call quality under load, Internet call quality (MOS), Voice-class latency, Jitter
DataDownload ↓, Upload ↑, Latency under load, Internet latency
RFSignal (dBm), SNR (dB), Downlink rate ↓, Band — AP data only

Home orders them Voice, Data, RF and starts cubes on Call quality (MOS); Professional orders them RF, Voice, Data. See Cube colors, labels, and detail.

Why did my cubes switch to Internet latency?

Section titled “Why did my cubes switch to Internet latency?”

If ARRF can’t time your router — none was found, or it never answers (for example when Local Network access was declined) — every Wi-Fi-hop cube would be grey. So, once per walk, if the colour is Wi-Fi latency, Jitter, Packet loss, or Voice-class latency, ARRF switches it to Internet latency, which is still measured.