UAV-Assisted Wireless Communications: An Experimental Analysis of Air-to-Ground and Ground-to-Air Channels in Open Environments

The study addresses how UAV heading and distance affect A2G and G2A wireless links in open, obstacle-free environments. Motivated by the need for rapid, on-demand connectivity (e.g., emergencies, crowd scenarios) and stringent QoS requirements (throughput, delay, PLR), the paper argues that accurate channel characterization is essential for reliable UAV-assisted communications. Prior work largely emphasizes A2G modeling without fully accounting for UAV heading, antenna orientation, or receiver height. The paper’s purpose is to experimentally characterize RSSI and TCP throughput versus distance and UAV heading, compare results against deterministic baselines (Friis and two-ray), and derive effective radiation patterns to inform system design and deployment.

The related work surveys A2G, G2A, G2G, and A2A channel measurements and models, including deterministic, stochastic, and geometric-stochastic approaches. Prior studies examined path loss, shadow fading, Doppler, PDP, RMS delay/Doppler spreads, and Rician K-factor across varied environments and altitudes. Measurements using LTE infrastructures investigated AoA and AS versus altitude and compared combining/beamforming with single-antenna systems. Additional studies explored receiver height effects over sea links, passive sounding (UMTS/GSM) up to 500 m altitudes, and empirical/urban path-loss models from 200 MHz to 5 GHz. Despite this body of work, the impact of UAV heading and real antenna orientation on RSSI/throughput in open-field A2G/G2A links remains underexplored, motivating this experimental characterization.

System setup: One UAV acts as a Wi‑Fi Access Point (AP) with LTE backhaul and one ground user equipment (UE, smartphone). The UAV carries a communications module with IEEE 802.11n (Wi‑Fi 4), MIMO 2x2, channel 1, 20 MHz bandwidth. Two 5 dBi omnidirectional Wi‑Fi antennas are horizontally mounted on the UAV payload: one oriented North/South and the other East/West (North is UAV head). Two vertical triband omnidirectional antennas provide LTE connectivity to a remote base station (BS). The UE is a Xiaomi Mi 9T smartphone (single antenna). Wi‑Fi auto-rate uses Minstrel; TCP downlink traffic (iperf3 server on UAV, client on smartphone) saturates the link; tcpdump on UAV records per-antenna RSSI. The LTE BS (Band 3, 1.8 GHz, no CA) is ~120 m from the hangar; theoretical DL/UL are up to 150/50 Mbit/s. Measured near-hangar LTE: ~114 Mbit/s DL and ~55 Mbit/s UL; at ~1.2 km: ~11 Mbit/s DL and ~1 Mbit/s UL. Experimental scenarios: (A) UAV at 50 m AGL, horizontal distance 200 m from the UE (Euclidean ~206 m). UAV heading is rotated in 45° increments through 360°. For each heading, per-antenna RSSI (dBm) at the UAV and UE downlink throughput (Mbit/s) are recorded. Effective radiation pattern (ERP) is derived and compared to Friis baseline. (B) Distance sweep at 50 m AGL: UAV moves away from the UE in 25 m steps with heading 180° (moving away), then returns via the same waypoints with heading 0° (coming back). For each waypoint, RSSI at UAV and UE downlink throughput are measured and compared to Friis and two‑ray models. (C) Backhaul evaluation: At ~1.42 km from the LTE BS, compare UE Internet throughput when directly connected to LTE vs. connected through the UAV (Wi‑Fi access plus LTE backhaul on the UAV).

- Antenna radiation and heading effects (Scenario A): The two horizontally mounted omnidirectional antennas exhibit heterogeneous effective radiation patterns due to mounting orientation and UAV body obstruction. Per-antenna RSSI versus heading deviates from the Friis isotropic baseline; summing both antennas mitigates deep nulls and yields a more uniform pattern. - Throughput vs. heading (Scenario A): Downlink throughput varies substantially with UAV heading, from roughly 30 Mbit/s at cardinal headings (N, S, E, W) to about 10 Mbit/s at diagonal headings (NE, SE, SW, NW), despite omnidirectional antenna specifications. RSSI minima near 180° are attributed to UAV body obstruction. - Distance dependence and rate adaptation behavior (Scenario B): RSSI is lower when the UAV moves away (heading 180°) than when it returns (heading 0°), consistent with body obstruction. When returning, throughput can be temporarily lower due to Minstrel’s delayed rate increase; restarting the Wi‑Fi card or pausing can restore optimal rates. A pronounced throughput drop occurs due to A2G/G2A link asymmetry (~10 dB): ACKs are forced at 24 Mbit/s or above; when G2A SNR cannot sustain ACKs at 24 Mbit/s, data rates are lowered despite adequate A2G SNR. - Coverage and throughput envelopes: For single‑antenna smartphones, green region (10–80 Mbit/s) radius is ~400 m along N/S/E/W and ~200 m along diagonals; blue region (>1 Mbit/s) radius is ~1100 m along N/S/E/W and ~550 m along diagonals. Internet connectivity was maintained beyond 1.5 km, and with a two‑antenna smartphone, a ~1500 m coverage radius was achieved in tests. - Backhaul via UAV vs. direct LTE (Scenario C): At the same obstructed location (~1.42 km from BS), direct LTE to the UE achieved up to ~13 Mbit/s, while via the UAV Wi‑Fi + LTE backhaul achieved up to ~21 Mbit/s (~1.6× gain), attributed to restoring radio LoS through the UAV. - Model comparison: Measured RSSI departs from deterministic Friis and two‑ray baselines due to real antenna patterns and UAV body effects; the experiments provide a more accurate characterization under these conditions.

The results demonstrate that UAV heading and physical configuration (antenna orientation and body obstruction) materially affect A2G/G2A link quality and throughput, even in obstacle‑free environments. This addresses the research gap by quantifying heading‑dependent deviations from deterministic models and by showing that summing signals from orthogonally oriented antennas reduces pattern nonuniformities. Practical implications include: (i) heading‑aware planning and antenna placement to avoid obstruction at critical headings (e.g., 180°), (ii) accounting for link asymmetry and ACK rate constraints in rate control policies, and (iii) leveraging UAV relays to restore LoS and improve end‑to‑end throughput in obstructed backhaul scenarios. The findings suggest deterministic models (Friis/two‑ray) are insufficient alone; empirical characterization incorporating heading and mounting effects is necessary for accurate planning and QoS assurance.

The study experimentally validates long‑range UAV‑assisted connectivity between a UAV and a ground UE, maintaining service up to ~1.5 km and achieving substantial throughput under Wi‑Fi 802.11n with LTE backhaul. It reveals heterogeneous effective radiation patterns and significant heading‑dependent performance variations due to antenna orientation and UAV body obstruction, with minima observed near 180°. Summing the two orthogonally mounted antennas improves uniformity. Distance sweeps highlighted rate adaptation dynamics (Minstrel lag) and throughput drops driven by A2G/G2A asymmetry and ACK rate thresholds. Using the UAV as a relay restored radio LoS and improved throughput by ~1.6× compared with direct LTE in a non‑LoS location. Future work includes deeper investigation of UAV motion dynamics on signal strength and throughput, improved antenna designs and placements to reduce heading sensitivity, and rate control mechanisms robust to link asymmetry.

- Environment: Tests were conducted in obstacle‑free, interference‑free open areas; results may not generalize to urban, cluttered, or high‑interference settings. - Hardware and bands: Single UAV platform, specific antenna mounts, 2.4 GHz 802.11n with Minstrel, and a specific LTE Band 3 BS; performance may differ with other bands, MCS sets, or rate control algorithms. - UE diversity: Primary UE was a single‑antenna smartphone (worst‑case); multi‑antenna UEs can significantly improve performance and coverage. - Geometry: Most characterizations used 50 m AGL and distances up to ~500 m for detailed A2G/G2A analysis; other altitudes and terrains were not exhaustively explored. - Model scope: Comparisons were to Friis and two‑ray baselines; broader model sets and statistical fading parameters (e.g., K‑factor, delay spreads) were not estimated here.