Wi-Fi 7 · Field Study
Industrial Project · BE (Hons) · Auckland University of Technology · 2025–2026

What affects Wi‑Fi 7 most?

A field study of the factors driving indoor throughput and signal strength. 779 iperf3 measurements across 76 configurations, varying band, channel width, distance, obstruction material and operating mode, using only consumer Wi-Fi 7 hardware in real homes and offices.

Authors Joshua Gudgin · Zoraver Malhotra Supervisor Professor Nurul Sarkar Runs 779 across 76 scenarios

§ 01 BackgroundWhat this study is, and why it exists

Wi-Fi 7 (IEEE 802.11be) advertises peak rates up to 46 Gbps, but vendors will only tell you that real-world speeds may vary. This study replaces that vague caveat with a ranked, evidence-based account of which factors actually move indoor Wi-Fi 7 throughput, and by how much.

The headline features of Wi-Fi 7 are 320 MHz channels, 4K-QAM modulation, and Multi-Link Operation (MLO). The hardware used here, an ASUS RT-BE92U access point with TP-Link Archer TBE400UH adapters on two Windows 11 laptops, characterises a consumer 160 MHz-class system rather than the theoretical Wi-Fi 7 maximum, because the TBE400UH uses enhanced multi-link single radio (eMLSR) and offers no 320 MHz channel.

Across 779 individual iperf3 runs over 76 configurations, throughput and signal strength were recorded while varying five layout factors: environment (four indoor settings), band (2.4 / 5 / 6 GHz), mode (Standard vs MLO), distance (1.5 m to 12 m), and obstruction material and count. The result is a measured picture of how consumer Wi-Fi 7 behaves in the conditions people actually use it.

779
iperf3 runs
76
Distinct scenarios
5
Layout factors varied
160MHz
Max channel width tested

§ 02 ScenariosBrowse all 76 test scenarios

Filter by environment, band, mode, or connection. Click any row to expand and see mean throughput, standard deviation, range, link rate, PHY efficiency, and interference metrics.

Reading the scenario table

  1. Scenario name tells you the location, what's in the path, the band, and the distance, e.g. Home-1×Plasterboard 5GHz 8m means 8 metres from the router through one plasterboard wall.
  2. Italic description below the name gives a plain-English sketch of where the router and client sat and what the signal had to pass through.
  3. Tag chips mark non-default settings: MLO for Multi-Link Operation, Wired for runs where the iperf3 server is cabled to the router (one wireless hop, isolating the Wi-Fi link), plus channel width and stream count. Without the Wired chip the server is also wireless, two endpoints sharing one radio, halving the available airtime.
  4. The bar shows mean throughput scaled against the maximum measured across the whole study (~1500 Mbps). Red bars indicate sub-100 Mbps results; green bars indicate gigabit-and-above.
  5. The number on the right is the mean iperf3 throughput in Mbps across all runs of that scenario.
  6. Click any row to expand and reveal standard deviation, min/max range, PHY link rate, PHY efficiency, co-channel and adjacent-channel interference, neighbour AP count, primary and secondary RSSI in dBm where measured, and for MLO, secondary-link details.
Scroll to explore all scenarios ↓

§ 03 SummaryHeadline findings

Indoor Wi-Fi 7 throughput was driven chiefly by frequency band and channel width, with obstruction material close behind. Distance mattered less than expected. Signal strength followed a different logic and was almost uncorrelated with throughput.

Steel floor
−87.0%
621 → 81 Mbps
Largest single-factor drop in the controlled comparisons.
2.4 GHz vs 6 GHz
−83.3%
592 → 99 Mbps
Band was the strongest predictor across the whole dataset.
Multi-Link Operation
−56.8%
593 → 256 Mbps
MLO lowered throughput in this consumer setup.
Wall present
−13.6%
671 → 580 Mbps
A single non-steel wall was a modest penalty.
Distance (1.5 m → 12 m)
−20.0%
99 → 79 Mbps
Weaker than expected, band and width dominate.
RSSI ↔ throughput
r ≈ −0.1
Signal bars are a poor proxy for real speed in Wi-Fi 7.
IThroughput vs RSSI

Signal strength and throughput are not the same thing

If this study has a single headline, it is that throughput and signal strength are driven by separate mechanisms. Throughput tracked the available spectrum, band and channel width, with RSSI tracking frequency-dependent path loss and obstruction material. The two correlated at only r ≈ -0.1.

The fastest connections in the dataset were often 6 GHz links with comparatively weak RSSI, while several stronger-RSSI links on narrower bands were slower. A strong signal alone is no guarantee of high throughput.

Consumer devices show signal bars instead of throughput but in Wi-Fi 7, the bars do not predict the speed.
IIBand & width

Band and channel width dominate everything else

Mean throughput rose from 108 Mbps on 2.4 GHz to 578 Mbps on 6 GHz, and from 74 Mbps on 20 MHz channels to 565 Mbps on 160 MHz. The widest channels exist only in the higher bands, so the two factors compound. This was the largest practical effect in the study.

But the band result interacted with obstruction. In open line-of-sight, 6 GHz averaged 664 Mbps against 399 Mbps for 5 GHz. Through the steel floor, the ranking reversed: 5 GHz sustained 133 Mbps while 6 GHz fell to 81 Mbps, because the higher-frequency signal attenuated more heavily through dense material.

The best band depends on what sits in the path. 6 GHz wins clean line-of-sight; 5 GHz wins through steel.
IIIMLO surprise

Multi-Link Operation made things slower

One of Wi-Fi 7's most heavily advertised features came out lower on average than ordinary single-link operation. MLO configurations averaged 207 Mbps against 393 Mbps for Standard, even though MLO ran eight parallel TCP streams against Standard's four.

The likely contributors include the TBE400UH adapter's enhanced multi-link single radio design (true simultaneous transmit-receive on two bands isn't supported), the overhead of managing two links through one access point, and Windows reporting behaviour. The finding aligns with prior work treating MLO performance as configuration-dependent rather than guaranteed.

Switching MLO on is no guarantee of more speed, especially with a single-radio client.

A predictive model confirms the ranking

A Random Forest trained on the five layout factors: band, mode, distance, obstruction material, obstruction count predicted throughput with a cross-validated R² of 0.693 ± 0.205 and a mean absolute error of 91.9 ± 32.7 Mbps, far surpassing a log-distance physics baseline (R² = −0.65). The model's feature importances reproduced the same factor ranking obtained from the direct controlled comparisons, lending independent support to the result.

The model was then exported to JavaScript and deployed as SignalApp, a browser-based tool for predicting throughput from a sketched room layout.

0.693R²
Cross-validated
±92Mbps MAE
Mean absolute error
5
Input features

Try the throughput predictor

SignalApp is the trained Random Forest above, made interactive. Sketch a room, place a router, drop a client pin, indicate the walls and get a predicted throughput in Mbps. The tool reports a ±92 Mbps uncertainty band rather than a single exact figure, reflecting what the model can honestly support.

Open the tool →