Network Performance Evaluation of Wi-Fi 6 and 5G in Mobile Deployments, and of 5G SA under Controlled Conditions

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

University of Waterloo

Abstract

Reliable wireless connectivity is important for mobile robots that use remote supervision, teleoperation, sensor-data transfer, and coordination with external computing resources. Performance during motion, however, may vary with trajectory geometry, obstruction, traffic direction, sustained load, infrastructure placement, radio conditions, and the complete end-to-end network path. This thesis evaluates these effects through three experimental campaigns: an outdoor trajectory-based evaluation of Wi-Fi 6 and public 5G Non-Standalone (NSA), a repeated indoor evaluation of Wi-Fi 6 and 5G NSA across two propagation geometries, and a controlled 5G Standalone (SA) attenuation experiment. The campaigns progress from realistic field observation toward a controlled examination of the relationship among radio quality, throughput, and latency under load. The outdoor campaign evaluated Wi-Fi 6 and public 5G NSA across eight repeated traversals of a nominal 97 m route using a controller-limited commanded speed of 1 m/s. Under no load, the single-access-point Wi-Fi deployment maintained 0% packet loss and average round-trip times between 4.9 and 7.0 ms, although maximum RTT values ranged from 91.3 to 126.0 ms and P99 jitter exceeded 50 ms in seven of the eight runs. Under sustained traffic, Wi-Fi RTT varied along the trajectory and included observations in the low-, intermediate-, and elevated-latency reference bands. Downlink throughput was generally higher and less variable across runs than uplink throughput, while the maximum throughput observed during motion remained below the corresponding stationary averages in both directions. In the repeated 5G NSA measurements, mean downlink throughput generally increased from approximately 140–210 Mbps in the earlier route portion to approximately 220–260 Mbps over parts of the later portion, broadly coinciding with improved LTE serving-cell SINR. Uplink throughput remained comparatively stable at approximately 70–87 Mbps despite larger SINR fluctuations. The loaded RTT patterns for both technologies were consistent with end-to-end standing queueing under sustained traffic, although the bottleneck location and contributing lower-layer mechanisms were not measured. A separate exploratory 5G run showed that modem-reported serving-cell PCI transitions could coincide temporally with substantial throughput reductions, but the single run did not establish their cause or repeatability. The 75 ms value adopted from prior robotic-control work was used only as a descriptive reference because it applies to one-way application-layer delay, whereas this campaign measured RTT. The indoor campaign evaluated twenty retained forward runs for each technology, trajectory, and traffic condition over common motion-aligned intervals of 57.00 s for Trajectory 1 and 50.00 s for Trajectory 2. Trajectory 1 progressed from an initially favorable region into corridor-induced NLOS, whereas Trajectory 2 provided an NLOS-to-LOS progression for the flexibly positioned Wi-Fi access point but remained predominantly under open-area inter-floor NLOS for the fixed 5G deployment. Trajectory-resolved throughput, P50 latency, and radio measurements showed clear local changes associated with these route geometries, but the run-level P99 results demonstrated that no trajectory was uniformly better across all metrics. For Wi-Fi 6, mean run-level downlink P99 latency was 384.359 ± 225.373 ms in Trajectory 1 and 1553.995 ± 403.559 ms in Trajectory 2, whereas uplink P99 remained much lower at 89.627 ± 4.844 and 85.586 ± 6.174 ms, respectively. For 5G NSA, mean downlink P99 decreased from 200.281 ± 66.629 ms in Trajectory 1 to 71.192 ± 7.679 ms in Trajectory 2, while mean uplink P99 decreased from 347.472 ± 81.972 to 214.930 ± 33.576 ms. The 5G uplink latency distributions require additional qualification because only 85.031% and 71.771% of scheduled probes received matched responses in Trajectories 1 and 2, respectively. The indoor results therefore distinguish typical trajectory-resolved behavior, upper-tail latency, run-to-run variability, and response availability rather than reducing performance to a single latency value or technology ranking. The controlled 5G SA campaign used a private srsRAN/Open5GS testbed and a direct conducted RF path with nominal 10, 20, 30, and 40 dB attenuator combinations. Five approximately five-minute runs were conducted for each attenuation and traffic condition. Central no-load RTT remained stable, with the mean of the run-level mean RTT values ranging only from 29.26 to 29.66 ms. Under sustained downlink load, mean modem-reported SINR decreased from 24.96 to 11.35 dB across the tested conditions, while mean throughput decreased from 51.76 to 12.19 Mbps. The mean of the run-level mean RTT values increased from 96.16 to 1078.70 ms, and the mean run-level P99 increased from 340.46 to 2223.06 ms. The largest throughput reduction and the first clear increase in loaded RTT occurred between the tested nominal 20 and 30 dB configurations. However, the attenuation values were not independently calibrated, the 10 and 20 dB loaded-latency ranges overlapped, and no intermediate configurations were evaluated. The measurements therefore do not establish a precise path-loss threshold or distinct operating regimes. The simultaneous throughput reduction and loaded-RTT increase are consistent with standing queue delay as the achievable service rate decreased, but queue occupancy, scheduler behavior, retransmissions, and other lower-layer mechanisms were not recorded. Together, the content chapters show that mobile-robot wireless performance depends on the evaluated trajectory, propagation geometry, traffic direction, offered load, radio condition, infrastructure placement, response availability, and complete end-to-end network path. Neither nominal connectivity nor route-wide average performance is sufficient to characterize all operating conditions, and the evaluated deployments do not support a universal Wi-Fi-versus-5G ranking. Deployment assessment should therefore combine trajectory-resolved throughput and typical latency with run-level upper-tail latency, probe-response coverage, run-to-run variability, and technology-appropriate radio measurements.

Description

Keywords

Citation

Endorsement

Review

Supplemented By

Referenced By