ReDrone5G: Field Trial Evaluation of Remote UAV Media and Control over Public and Private 5G Networks

ReDrone5G: Field Trial Evaluation of Remote UAV Media and Control over Public and Private 5G Networks Kashif Mehmood∗ , Tu Dac Ho∗ , Katina Kralevska∗ , Hacı Mehmet Güzey†

∗ Norwegian University of Science and Technology (NTNU), Trondheim, Norway

Email: {kashif.mehmood, tu.d.ho, katinak}@ntnu.no † Sivas University of Science and Technology; Ada Guzey Engineering, Sivas, Türkiye

arXiv:2611.07068v1 [cs.NI] 4 Nov 2026

Email: mehmet.guzey@sivas.edu.tr Abstract—Drone-assisted live media production limited radio and core-network information. Private 5G stresses cellular networks by combining high-rate up- deployments provide greater access to configuration and link video, continuous telemetry, and latency-sensitive measurement, but do not reproduce all sources of variability command/control on a mobile aerial platform. This paper presents ReDrone5G, an end-to-end field-trial present in commercial networks. This paper presents ReDrone5G, an end-to-end fieldplatform evaluated across two practical fifth-generation (5G) deployment paths: a commercial public non- trial platform for remote UAV media and control. The standalone (NSA) network in Sivas, Turkiye, and a platform integrates a drone-mounted camera, MAVLink private OpenAirInterface (OAI)-based standalone (SA) telemetry/control, a MediaMTX/WebRTC video pipeline, testbed in Trondheim, Norway. The same drone, media, telemetry/control, and monitoring stack is used in a ground-station control dashboard, and a monitoring stack. both trials, which are treated as complementary, non- The same application workflow is evaluated over two pracequivalent field deployments rather than a controlled tical but non-equivalent deployments: a commercial public A/B comparison. Results show that the commercial 5G non-standalone (NSA) network in Sivas, Türkiye, and deployment sustains remote drone media operation, an OpenAirInterface (OAI)-based private 5G standalone while the evaluated private deployment exhibits lower control RTT and video latency and higher video through- (SA) testbed in Trondheim, Norway. The deployments differ put. The private setup also exposes signal-to-noise in architecture, geography, radio conditions, network path, ratio (SNR), reference signal received power (RSRP), available bandwidth, network load, and, in the public case, modulation and coding scheme (MCS), physical resource the use of a virtual private network (VPN) overlay. The block (PRB), and block error rate (BLER), providing study therefore characterizes deployment-level behavior cross-layer radio/network context. ReDrone5G shows that public 5G is useful for feasibility validation, while rather than attributing measured differences to a single private 5G enables repeatable experimentation, tuning, public/private or NSA/SA factor. and future service assurance for drone-media workflows. Accordingly, the evaluation addresses two questions:

RQ1: Can an integrated UAV video, telemetry, and control workflow be sustained over commercial 5G under practical I. Introduction flight conditions? Drone-assisted live media production is relevant to sports RQ2: What additional performance characterization and and cultural events, infrastructure documentation, and radio/network observability become available when the remote production workflows. Unlike conventional best- same workflow is evaluated on a controllable private 5G effort video upload, this application combines three hetero- SA testbed? geneous traffic classes on a mobile aerial platform: high-rate This paper addresses these questions through the followuplink video, continuous telemetry, and latency-sensitive ing contributions: command/control. Its performance therefore depends not • We present an end-to-end UAV media and control only on connectivity, but also on whether the network field-trial platform combining live video, MAVLink can sustain predictable application behavior while the telemetry/control, a ground-station interface, and unmanned aerial vehicle (UAV) is moving. application/network monitoring. Fifth generation (5G) is a natural connectivity option • We evaluate the same application workflow over because it combines wide-area access with managed quality commercial 5G NSA and private OAI-based 5G SA of service (QoS) and support for differentiated traffic deployments, including simultaneous PDU sessions through 5G QoS identifier (5QI) profiles, protocol data with distinct single network slice selection assistance unit (PDU) sessions, and network slicing [1]. However, informations (S-NSSAIs) in the private network. the experimental capabilities depend strongly on the • We characterize application-level latency, jitter, and deployment model. Commercial networks provide realistic throughput across both deployments and complement coverage and operator-managed infrastructure, but expose the private-network measurements with radio-layer

TABLE I: Positioning of ReDrone5G relative to representative experimental UAV-over-5G and open 5G studies. “P” denotes partial coverage, while “–” indicates that the capability is not established by the scope discussed in the cited work. Work Muzaffar et al. [2] Baltaci et al. [3]

Scope / network context

Real UAV

Early experimental characterization of a 5G-connected drone. Real-time cellular video delivery for remotely piloted aerial vehicles.

Yes

–

–

–

–

–

P

P

Yes

P

–

Yes

–

–

Yes

–

Yes

Yes

–

–

–

–

Yes

–

Yes

P

P

P

P

P

Yes

P

Yes

–

Yes

Yes

–

–

Yes

P

–

–

Yes

–

–

–

–

Yes

Yes

Yes

Yes

Yes

Yes

Yesa

Yes

Yes

Candal-Ventureira et UAV computation offloading over a carrier-grade al. [4] 5G operator network. 5G!Drones [5] Large-scale trials of multiple UAV vertical applications over advanced 5G facilities. Bertizzolo et al. [6] Drone-sourced video streaming over a programmable 5G Open-RAN testbed. Performance evaluation of an open-source 5G SA Sahbafard et al. [7] deployment. ReDrone5G

a

Integrated UAV video and MAVLink telemetry/control across commercial 5G NSA and private OAI-based 5G SA deployments.

Public Private 5G 5G

UAV Video

Control/Network App. Telem. Slicing KPIs

RAN KPIs

Two simultaneous PDU sessions with distinct S-NSSAIs are configured for media and combined control/telemetry traffic; slice isolation under competing load is not evaluated.

key performance indicators (KPIs), exposing the additional diagnostic capability of an open experimental network. II. Related Work Research relevant to ReDrone5G spans three areas: cellular connectivity for aerial UAVs, application-level UAV media and control over 5G, and programmable private 5G platforms for experimentation. A. Cellular Connectivity and UAV Applications

processing and control functions to be placed closer to users [12, 13]. Large trial initiatives such as 5G!Drones have validated multiple UAV vertical applications over advanced 5G facilities [5]. Bertizzolo et al. [6] further demonstrated drone-sourced video streaming over a programmable 5G Open-RAN platform. These works establish the feasibility of UAV connectivity, media delivery, edge processing, and programmable radio experimentation, but typically investigate individual application or network aspects rather than an integrated workflow for media, telemetry, and control.

UAV communication research has extensively studied aerial propagation, interference, coverage, mobility, trajec- B. QoS and Private 5G Experimentation The heterogeneous UAV workload makes the 5G QoS tory optimization, and the use of UAVs as aerial users or communication infrastructure [8, 9, 10, 11]. These framework particularly relevant. The 5G system supports studies show that UAV altitude and mobility create radio differentiated traffic through QoS flows, standardized 5QI conditions that differ from conventional terrestrial-user characteristics, PDU sessions, policy control, and network operation. Their primary focus, however, is generally the slicing [1]. These mechanisms allow traffic classes with wireless link or network architecture rather than a com- different performance objectives to be handled separately; plete remote UAV application. Experimental studies have video emphasizes sustained uplink throughput, whereas complemented this work with real cellular deployments. telemetry and command/control emphasize timely packet Muzaffar et al. [2] demonstrated practical 5G-connected delivery. The extent to which these mechanisms can be UAV operation, exposing effects that are difficult to studied depends on the deployment environment. Commercapture through simulation alone. For drone-assisted media cial networks provide realistic operator-managed coverage, production, however, connectivity must support three routing, and shared-network conditions, but external users concurrent traffic classes: high-rate uplink video, continuous generally have limited access to scheduler configuration, QoS mapping, user plane function (UPF) placement, radiotelemetry, and latency-sensitive command/control. Application-oriented studies have investigated parts of resource allocation, and core-network routing. this workload. Baltaci et al. [3] analyzed real-time video Private 5G networks provide complementary capabilities delivery over cellular networks for remote piloting, where through local control, data locality, and service customizamedia delay directly affects operator situational awareness. tion [14, 15, 16]. Open-source implementations further Candal-Ventureira et al. [4] evaluated UAV computation expose configuration and measurement interfaces needed for offloading in carrier-grade 5G, showing how network path repeatable experimentation. Sahbafard et al. [7] evaluated and edge placement affect service performance. More gen- an open-source 5G SA deployment, while OAI provides erally, multi-access edge computing (MEC) enables media open implementations of both the radio access network

Fig. 1: ReDrone5G experimental architecture for the commercial 5G NSA deployment in Sivas, Türkiye, and the private OAI-based 5G SA deployment in Trondheim, Norway.

(RAN) and core network [17]. Such platforms allow appli- and exposes signal-to-noise ratio (SNR), reference signal cation observations to be examined alongside radio and received power (RSRP), modulation and coding scheme network indicators that are normally unavailable through (MCS), physical resource block (PRB) allocation, and block commercial access. This additional visibility supports cross- error rate (BLER). This provides radio/network context for layer interpretation, although it does not by itself establish application measurements and a basis for future controlled causal relationships. studies of slice isolation and closed-loop service assurance. Table I summarizes ReDrone5G’s positioning relative to C. Positioning of ReDrone5G representative experimental UAV-over-5G studies. Existing work therefore addresses important components III. ReDrone5G System Design of the problem: cellular UAV connectivity [2], remotepiloting video [3], UAV computation over operator 5G [4], A. End-to-End Architecture large-scale UAV trials [5], and programmable Open-RAN ReDrone5G integrates the airborne UAV, cellular access video streaming [6]. ReDrone5G is positioned at the network, and ground station into a common experimental intersection of these directions. workflow, as shown in Figure 1. The airborne platform Three gaps motivate this work. First, prior studies typi- generates video and MAVLink telemetry/control traffic, cally emphasize a single dominant UAV service component, while the ground station receives video, monitors flight, and whereas ReDrone5G evaluates uplink video, MAVLink supports operator interaction. The application workflow telemetry, and command/control concurrently during real is the same across public and private deployments; the UAV operation. Second, commercial and experimental main differences are the underlying network architecture, 5G environments provide different forms of evidence but routing path, and available observability. are rarely evaluated using the same UAV application workflow. ReDrone5G therefore uses the same airborne B. Private 5G SA Deployment The private deployment uses an open-source OAI 5G media, telemetry/control, ground-station, and monitoring stack over a commercial 5G NSA network and an OAI- SA RAN and core network. The gNB runs with a USRP based private 5G SA network. Because the deployments B210 software-defined radio (SDR) on band n77 using differ in architecture, geography, radio configuration, rout- 20 MHz bandwidth and 30 kHz SCS. The RAN, CN, UPF, ing, load, and VPN usage, we treat them as comple- monitoring infrastructure, and ground station are hosted mentary field settings rather than a controlled public- within the controlled experimental environment. versus-private benchmark. Third, application performance Traffic was separated into two service classes using two in commercial networks is difficult to relate to internal simultaneous PDU sessions established by the airborne network behavior. The private ReDrone5G deployment uses user equipment (UE), with each session associated with two simultaneous PDU sessions with distinct S-NSSAIs, a distinct S-NSSAI. The media session carried the uplink separating media from combined control/telemetry traffic, video flow using 5QI 7, while the second session carried

TABLE II: Configuration and measurement scope of the two ReDrone5G field deployments. Parameter

Commercial 5G NSA

Private 5G SA

Location Network Radio configuration

Sivas, Türkiye Turkcell commercial infrastructure Operator managed; detailed configuration unavailable

Traffic differentiation

Operator controlled; internal mapping unavailable

Network path

Commercial network, public routing, and Tailscale/WireGuard overlay Video, telemetry, control Not exposed

Trondheim, Norway Open-source OAI RAN and 5G core Band n77, 20 MHz, 30 kHz subcarrier spacing (SCS), universal software radio peripheral (USRP) B210 Two simultaneous PDU sessions with distinct S-NSSAIs; 5QI 7 for media and 5QI 5 for control/telemetry Local controlled RAN/core network (CN)/UPF path Video, telemetry, control SNR, RSRP, MCS, PRB, BLER, and network configuration

Application observability Radio/core observability

control and telemetry traffic using 5QI 5. Both sessions TABLE III: ReDrone5G application-level KPI targets used remained active concurrently during the integrated trials. in the evaluation. The selected 5QI values are standardized non-guaranteed KPI Flow 5QI Target bit rate (NGBR) profiles [1]; we use them here as practical Control 5 < 150 ms traffic mappings rather than as optimal configurations Control RTT Control jitter Control 5 < 50 ms for safety-critical control. The experiments demonstrate Video 7 < 100 ms concurrent session establishment and service-class separa- Video latency Video jitter Video 7 < 20 ms tion, but do not claim slice-isolation or prioritization gains Video Video 7 ≥ 10 Mbps because no controlled competing traffic was introduced. throughput The private deployment exposes RAN and CN configu- Telemetry Telemetry 5 ≥ 0.05 Mbps ration, routing, and monitoring information. Application throughput measurements can therefore be examined together with IV. Experimental Methodology radio indicators such as SNR, RSRP, MCS, PRB allocation, and BLER. These measurements provide cross-layer We evaluated ReDrone5G through a structured threecontext; causal relationships require additional controlled phase trial program that progressively validated the airor time-aligned analysis. borne hardware, application stack, and cellular-network integration. All phases used the same application stack: C. Commercial 5G NSA Deployment MediaMTX-based video streaming, MAVLink telemeThe public deployment uses Turkcell’s commercial 5G try/control, and Prometheus/Grafana observability backed NSA network in Sivas, Türkiye. A Tailscale/WireGuard by InfluxDB. The private 5G SA trials were conducted in VPN overlay provides stable bidirectional reachability Trondheim, Norway, using the OAI-based infrastructure between the airborne UE and the ground station across described in Section III-B. The public 5G NSA trials were carrier NAT and public-network routing boundaries. The conducted in Sivas, Türkiye, over Turkcell’s commercial private deployment does not require this overlay because network. Table III defines the target KPIs we use to assess its radio, core, UPF, and ground-station components are the results. located within the same controlled network. Consequently, the public measurements characterize A. Experimental Trials the complete commercial-access path, including 5G NSA, The evaluation followed three phases. In Phase 1, indoor operator and Internet routing, and VPN encapsulation. and outdoor tests using direct radio control (RC) and The overlay is therefore a confounding factor rather than Wi-Fi links validated the airborne platform, including the part of a pure radio-access measurement. Its encryption, Pixhawk Cube, DJI Action 4 camera, and cellular UE, relay selection, NAT traversal, and endpoint processing before cellular integration. may contribute to the observed latency or throughput. In Phase 2, the private OAI deployment validated two Only application-level measurements are available from simultaneous PDU sessions and established controlled the commercial deployment. Scheduler behavior, QoS indoor application baselines. One airborne UE was used enforcement, UPF placement, PRB allocation, MCS, and without competing UEs or controlled background traffic. other radio/core indicators remain under operator con- Consequently, these experiments validate concurrent sestrol. Performance changes can therefore be observed but sion operation and traffic separation, but not slice isolation cannot be attributed to a particular radio-, core-, routing- or prioritization under congestion. , or overlay-layer mechanism. Table II shows a detailed Phase 3 evaluated the complete end-to-end (E2E) system comparison of the two setups. during field flights. The public trial used Turkcell 5G NSA

TABLE IV: Application-level KPI performance and target compliance for the public 5G NSA and private 5G SA deployments. Nw denotes the number of one-second analysis windows. KPI

Network

Mean ± SDa

Tail

Target compliance

Nw

Control RTT Control RTT

Public NSA Private SA

98 ± 10.9 ms 67.5 ± 4.2 ms

p95 = 150 ms p95 = 73 ms

≈ 95% < 150 ms > 99% < 150 ms

15976 18657

Control jitter Control jitter

Public NSA Private SA

11.2 ± 4.5 ms 24.3 ± 1.8 ms

p95 = 18 ms p95 = 28 ms

> 99% < 50 ms > 99% < 50 ms

15976 18657

Video latency Video latency

Public NSA Private SA

77.5 ± 15.7 ms 43.1 ± 3.8 ms

p95 = 105 ms p95 = 51 ms

≈ 93% < 100 ms > 99% < 100 ms

15976 18657

Video jitter Video jitter

Public NSA Private SA

5.8 ± 1.4 ms 5.0 ± 0.7 ms

p95 = 8.8 ms p95 = 7.5 ms

> 99% < 20 ms > 99% < 20 ms

15976 18657

Video throughput Video throughput

Public NSA Private SA

5.6 ± 1.3 Mbps 14.6 ± 0.9 Mbps

p5 ≈ 4.0 Mbps p5 ≈ 11.8 Mbps

< 1% ≥ 10 Mbps > 99% ≥ 10 Mbps

15976 18657

Telemetry throughput Telemetry throughput

Public NSA Private SA

0.08 ± 0.001 Mbps 0.115 ± 0.007 Mbps

p5 ≈ 0.055 Mbps p5 ≈ 0.085 Mbps

> 99% ≥ 0.05 Mbps > 99% ≥ 0.05 Mbps

15976 18657

a

SD denotes the standard deviation reported in the field-test summaries; entries without ± give the reported mean only. SD describes variability and is not a confidence interval.

(a) Orbital/spiral

(b) Horizontal passes

Fig. 2: UAV flight profiles during the private 5G SA trials in Trondheim: (a) orbital/spiral motion with altitude changes and (b) repeated horizontal passes.

with the Tailscale overlay at altitudes of approximately 5 m to 25 m and ranges up to 100 m. We evaluated the private deployment across two flight campaigns that included altitude changes and linear, orbital, and spiral trajectories. Unless otherwise stated, the reported distributions pool observations across the corresponding field campaigns. Figure 2 illustrates the orbital/spiral and horizontal flight profiles used during the private 5G SA trials in Trondheim.

deployment; the public commercial network did not expose equivalent RAN/core measurements to users. Radio-layer counters were sampled every 10 ms. Application and radio measurements were subsequently aggregated into one-second analysis windows for the reported distributions. The deployments constitute a comparative field study rather than paired or controlled A/B observations. They differ in architecture, geography, radio configuration, network load, routing, and observability; the public path also We assessed performance against the KPIs in Table III includes a VPN overlay. Reported differences therefore using metrics collected at the application, transport, and radio layers. Application-layer metrics comprised video represent the behavior of the evaluated deployment conthroughput and E2E latency, command/control round-trip figurations and should not be attributed solely to network time (RTT) and jitter, and telemetry delivery rate. Radio- ownership or NSA/SA architecture. layer indicators comprised SNR, RSRP, MCS, BLER, and PRB utilization. We computed confidence intervals using B. Measurement Methods bootstrap resampling on window-level KPI aggregates rather than individual samples to reduce the effect of We measured control RTT at the application layer as temporal correlation in the measurements. In the private the elapsed time from issuing a command at the ground OAI deployment, we collected radio-layer KPIs from the station to receiving the corresponding acknowledgment or OAI/FlexRIC monitoring path and exported them to state update via the MAVLink/control path. Control jitter the time-series monitoring stack. The collected indica- was extracted from MAVLink by tracking the variance tors included SNR, RSRP, MCS, PRB allocation, and in network delay over time using consecutive message BLER. These counters were available only in the private timestamps.

Fig. 3: Application-layer KPI CDFs for public 5G NSA and private 5G SA deployments. We measured video latency between frame generation Control RTT and jitter: The private 5G SA deployor availability in the drone-side media pipeline and frame ment exhibits a tighter control RTT distribution, with reception/rendering at the ground station. Endpoint timing a mean of 67.5 ms, p95 of 73 ms, and p99 of 76 ms. The was obtained using timestamping. The resulting metric public deployment has a mean of 98 ms, while its p95 is therefore reported as application-observed E2E media reaches the 150 ms application target and its p99 increases delay rather than pure radio one-way latency. We calculated to 175 ms. Thus, the public mean remains below the target, video jitter from Real-time Transport Protocol (RTP) inter- but its upper tail still contains violations. Control jitter arrival time statistics. is lower on the public-plus-overlay path (11.2 ms mean) Video throughput is the received application-layer media than on the private path (24.3 ms mean). Because no rate and therefore reflects the combined behavior of the controlled scheduler, traffic-load, or time division duplex encoder, MediaMTX/WebRTC pipeline, transport path, (TDD) ablation was performed, this difference is reported and cellular network rather than the maximum physical- descriptively rather than attributed to a particular network layer capacity. The video configuration was 1080p at 30 mechanism. FPS, using H.265 with a maximum bitrate of 25 Mbps; Video latency and jitter: Mean application-observed both deployments used the same application configuration. video latency decreases from 77.5 ms in the public deployRadio-layer measurements in the private deployment ment to 43.1 ms in the private deployment. Although the were collected through the OAI/FlexRIC monitoring path public mean is below the selected 100 ms target, its p95 and included SNR, RSRP, MCS, PRB allocation, and and p99 reach 105 ms and 118 ms, respectively. The private BLER. For the reported distributions, measurements were distribution remains tighter, with p95 of 51 ms and p99 of aggregated into one-second analysis windows. The sample 55 ms. Mean video jitter is low in both deployments (5.8 ms count reported in Table IV therefore denotes analysis public and 5.0 ms private). windows rather than independent flight repetitions. Video and telemetry throughput: Video throughput shows the largest separation between the evaluated configuV. Performance Evaluation and Results rations. The public distribution remains below the selected A. Application-Level Performance application target of 10 Mbps for most observations, with a Figure 3 presents six application-level distributions: mean of approximately 5.6 Mbps. The private deployment control RTT, control jitter, video latency, video jitter, has a mean of approximately 14.6 Mbps, and its distribution video throughput, and telemetry throughput. Table IV shifts toward higher throughput. This deployment-level summarizes the six metrics in terms of their central, tail, difference cannot be attributed solely to private-network and target-compliance behavior. resource availability, as radio conditions, architecture, net-

Fig. 4: Radio-layer KPI cumulative distribution functions (CDFs) for the private 5G SA deployment (Trondheim, Norway). work path, load, and VPN usage also differ between trials. as evidence of application-level latency or throughput Telemetry throughput remains above the selected low- differences. rate target in both deployments, indicating that telemetry MCS and PRB allocation: The downlink MCS distritransport was not the dominant application bottleneck bution occupies a higher range than the uplink distribution, under the evaluated conditions. while the uplink values are concentrated in the mid-range. Allocated PRBs vary from a small number to the available B. Private 5G SA Radio-Layer Behavior 51 PRBs, showing that radio-resource allocation changes Figure 4 presents radio-layer CDFs collected from the substantially over the trial. The observed MCS and PRB private 5G SA deployment. Equivalent measurements were distributions are consistent with dynamic link adaptation not exposed by the commercial network. The measurements and scheduling under varying radio and traffic conditions. therefore provide radio-layer context for the private field trial, although the marginal distributions alone do not C. Discussion and Limitations establish causal relationships with individual applicationPerformance improvements were metric-dependent. The level observations. private deployment exhibited lower mean control RTT SNR and RSRP: The uplink SNR distribution is concentrated primarily between approximately 12.5 dB and (67.5 versus 98 ms), lower video latency (43.1 versus 77.5 20 dB, with a mean near 15 dB. The RSRP distribution ms), and higher video throughput (14.6 versus 5.6 Mbps), is concentrated between approximately −95 dBm and but higher mean control jitter (24.3 versus 11.2 ms). Both −90 dBm. The variation across both indicators is consistent mean jitter values remained below the selected 50 ms target. with measurements collected during actual flight rather The private deployment therefore did not improve every reported metric. than under static radio conditions. BLER: Both downlink and uplink BLER distributions Successful operation also did not imply compliance with are concentrated at low values, with the uplink distribution every application target. The public deployment sustained extending farther to the right than the downlink distribu- the integrated workflow, while video throughput remained tion. These observations are consistent with a generally below the selected 10 Mbps threshold for most observations reliable radio link during the evaluated flight campaign. and upper-tail control RTT and video latency crossed Without time-aligned retransmission analysis, however, their respective thresholds. Mean performance alone would the BLER distribution alone should not be interpreted obscure these limitations.

These observations characterize the evaluated configurations. Architecture, geography, radio conditions, load, routing, and public-path VPN use differ between deployments, preventing causal attribution to network ownership or NSA/SA architecture. Private-network radio counters provide diagnostic context, but the reported marginal distributions do not establish application–radio causality. Two concurrent PDU sessions demonstrate traffic separation; slice isolation under competing load and trajectory-specific application effects remain unquantified. VI. Conclusion ReDrone5G demonstrated concurrent UAV video, telemetry, and command/control across commercial NSA and private OAI-based SA deployments. The public deployment sustained operation but did not consistently meet the selected video-throughput and latency targets. The private deployment provided lower mean latency, higher video throughput, and additional radio observability, while exhibiting higher control jitter. These deployment-specific results motivate evaluation across multiple KPIs and distribution tails. Future work will quantify uncertainty across repeated flights, trajectory-specific performance, and slice isolation under controlled traffic competition. Acknowledgments This work was supported by the ReDrone5G project under FIDAL (Field Trials Beyond 5G), Grant Agreement No. 101096146. The authors gratefully acknowledge the FIDAL consortium and project partners for supporting the field-trial infrastructure, experimentation, and validation activities presented in this paper. References [1] 3GPP, “TS 23.501: System Architecture for the 5G System,” tech. rep., 3rd Generation Partnership Project, 2024. Release 18. [2] R. Muzaffar, C. Raffelsberger, A. Fakhreddine, J. Lopez Luque, D. Emini, and C. Bettstetter, “First experiments with a 5g-connected drone,” in Proceedings of the ACM Workshop on Micro Aerial Vehicle Networks, Systems, and Applications, pp. 1–5, 2020. [3] A. Baltaci, D. Schupke, and V. Bajpai, “Analyzing realtime video delivery over cellular networks for remote piloting aerial vehicles,” in Proceedings of the ACM Internet Measurement Conference, 2022. [4] D. Candal-Ventureira, F. J. González-Castaño, F. GilCastiñeira, and P. Fondo-Ferreiro, “Is the edge really necessary for drone computing offloading? an experimental assessment in carrier-grade 5g operator networks,” arXiv preprint arXiv:2403.19729, 2024. [5] 5G!Drones Project, “D4.4 trial evaluation report: Unmanned aerial vehicle vertical applications’ trials leveraging advanced 5g facilities,” tech. rep., European Union Horizon 2020 5G!Drones Project, 2022. [6] L. Bertizzolo, T. X. Tran, J. Buczek, B. Balasubramanian, R. Jana, Y. Zhou, and T. Melodia, “Streaming

from the air: Enabling drone-sourced video streaming applications on 5g open-ran architectures,” in Proceedings of the ACM Workshop on Wireless Network Testbeds, Experimental Evaluation and Characterization, 2021. [7] A. Sahbafard, R. Schmidt, F. Kaltenberger, A. Springer, and H.-P. Bernhard, “On the performance of an indoor open-source 5g standalone deployment,” in 2023 IEEE Wireless Communications and Networking Conference (WCNC), pp. 1–6, 2023. [8] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless communications with unmanned aerial vehicles: Opportunities and challenges,” IEEE Communications Magazine, vol. 54, no. 5, pp. 36–42, 2016. [9] Y. Zeng, Q. Wu, and R. Zhang, “Accessing from the sky: A tutorial on uav communications for 5g and beyond,” Proceedings of the IEEE, vol. 107, no. 12, pp. 2327–2375, 2019. [10] M. Mozaffari, W. Saad, M. Bennis, Y.-H. Nam, and M. Debbah, “A tutorial on uavs for wireless networks: Applications, challenges, and open problems,” IEEE Communications Surveys & Tutorials, vol. 21, no. 3, pp. 2334–2360, 2019. [11] G. Geraci, A. Garcia-Rodriguez, D. Lopez-Perez, E. Bjornson, B. Galkin, L. Galati-Giordano, and M. Zorzi, “What will the future of uav cellular communications be? a flight from 5g to 6g,” IEEE Communications Surveys & Tutorials, vol. 24, no. 3, pp. 1304–1335, 2022. [12] T. Taleb, K. Samdanis, B. Mada, H. Flinck, S. Dutta, and D. Sabella, “On multi-access edge computing: A survey of the emerging 5g network edge cloud architecture and orchestration,” IEEE Communications Magazine, vol. 55, no. 6, pp. 64–70, 2017. [13] P. Mach and Z. Becvar, “Mobile edge computing: A survey on architecture and computation offloading,” IEEE Communications Surveys & Tutorials, vol. 19, no. 3, pp. 1628–1656, 2017. [14] A. Aijaz, “Private 5g: The future of industrial wireless,” IEEE Industrial Electronics Magazine, vol. 14, no. 4, pp. 136–145, 2020. [15] J. Ordonez-Lucena, J. Folgueira Chavarria, L. M. Contreras, and A. Pastor, “The use of 5g non-public networks to support industry 4.0 scenarios,” in 2019 IEEE Conference on Standards for Communications and Networking (CSCN), pp. 1–7, 2019. [16] 5G Alliance for Connected Industries and Automation, “5g non-public networks for industrial scenarios,” tech. rep., 5G-ACIA, 2019. [17] OpenAirInterface Software Alliance, “Openairinterface 5g radio access network and core network.” https:// openairinterface.org/, 2026. Accessed 2026.

Related documents

• OmniDroneX: An LLM-Assisted Holistic Drone-as-a-Service Ecosystem#282856
• Enabling Beyond-Visual-Line-of-Sight Drones Operation over Open RAN 5G Networks with Slicing#441725
• Performance Analysis of Tri-Sector Reflector Antennas for HAPS-Based Cellular Networks#1458275
• Farmers open to using drones for efficient planting#245183
• HAP-Centric Flying Ad-Hoc Networks With Cell-Free Non-Terrestrial Connectivity#662217