Title: Unveiling Low-Altitude 5G Performance: Linking Key Influencing Factors with UAV Flight Parameters
Authors: Xinzhe Liu (Pengcheng Laboratory, Shenzhen, China; South China University of Technology, Guangzhou, China); Jianer Zhou* (Pengcheng Laboratory, Shenzhen, China); Xiaoyong Ni (Pengcheng Laboratory, Shenzhen, China); Ke Luo (Pengcheng Laboratory, Shenzhen, China); Zhenyu Li (Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China); Xiaofeng Tao (Beijing University of Posts and Telecommunications, Beijing, China; Pengcheng Laboratory, Shenzhen, China); Weichao Li* (Pengcheng Laboratory, Shenzhen, China).
Introduction
This paper investigates how commercial 5G networks perform for low-altitude UAVs and, more importantly, why their throughput and latency differ from ground users. Reliable aerial connectivity is essential for safety-critical drone control, video transmission, logistics, and remote piloting, yet existing 5G deployments are engineered primarily for ground devices. UAVs move in three dimensions, often receive side-lobe coverage, see many cells simultaneously, and experience line-of-sight-dominated channels rather than the rich multipath environment assumed by terrestrial designs. Prior work often reports only end-to-end throughput or latency, which makes conflicting observations difficult to interpret and does not reveal the lower-layer mechanisms responsible for poor aerial performance.
Key idea and contribution
The authors build a lightweight airborne cross-layer measurement platform mounted on an industrial UAV. It combines rooted commercial phones, Accuver XCAL for PHY/MAC/RRC logging, and iPerf3 for end-to-end throughput and RTT measurements. The campaign spans urban and rural routes in Shenzhen, four Chinese mobile operators, altitudes below 400 m, and flight speeds up to 15 m/s, producing about 300 GB of synchronized data. This design lets the authors connect application-level performance with cellular configurations, radio conditions, resource scheduling, and mobility signaling.
The central contribution is a mechanism-level explanation of low-altitude 5G behavior. The paper identifies four primary influencing factors: spatial layers (SLs), modulation order (MO), resource-block (RB) allocation, and handovers (HOs). Altitude does not simply “cause lower throughput”; it changes path loss, line-of-sight dominance, visible-cell sets, and thus these four mechanisms. The authors also identify a “pseudo-successful handover,” where control-plane signaling reports success because the downlink remains reachable, but the UAV cannot complete uplink random access to the target cell. The device therefore appears connected while data transmission remains unavailable.
Evaluation
The evaluation compares ground and aerial performance across operators and then isolates altitude, speed, environment, time period, and frequency-band effects through controlled flights. Low-altitude UAVs consistently show lower uplink and downlink throughput and higher RTT than ground users; increasing altitude reduces usable spatial layers and pushes modulation toward more robust but less efficient settings, while also increasing handover frequency and failure risk. Urban and peak-load settings reduce and destabilize RB allocations, and pseudo-successful handovers are especially common under dense urban interference and evening load. The experiments also show that carrier aggregation can increase aerial downlink peak throughput to nearly 1 Gbps, although its dynamic activation can create latency spikes. This result is significant because it turns vague observations that “5G is unreliable in the sky” into concrete, measurable mechanisms that operators can target in network configuration, simulation, and handover design.
Personal thoughts
I like that the paper goes beyond reporting average throughput and latency. Its strongest aspect is the cross-layer causal story: flight parameters affect the channel and network state, which affect SL, MO, RB allocation, and HO behavior, which finally affect user-visible performance. The pseudo-successful handover finding is particularly compelling because it exposes a failure mode that ordinary end-to-end measurements could easily misclassify or miss. The controlled one-variable-at-a-time experiments also make the analysis more convincing than a purely observational flight study.
A limitation is that the measurements are concentrated in one city, one national market, and a limited set of commercial deployments, so the conclusions about particular bands and operator configurations may not generalize worldwide. It would be valuable to validate the findings across countries, antenna designs, rural macrocell deployments, and mmWave or 5G-Advanced networks. I would also like to see a prototype of the proposed bidirectional access check before handover, plus an evaluation of its overhead and its effect on false handovers, outage time, and drone-control quality. More broadly, this paper suggests that aerial networking needs mobility policies designed explicitly for 3D devices rather than small adjustments to terrestrial defaults.