Unmanned Aerial Vehicles (UAVs) are increasingly used in various applications due to their cost-effectiveness, maneuverability, and ability to provide on-demand wireless connectivity. 5G and beyond cellular networks envision UAVs as a solution for quickly deploying temporary wireless networks in emergencies or crowded areas. However, establishing reliable wireless links with UAVs presents challenges. Different QoS requirements exist depending on the type of data being transmitted (e.g., non-critical video vs. critical control signals). Accurate wireless channel modeling is crucial for optimizing system design and ensuring reliable broadband wireless links considering throughput, Packet Loss Ratio (PLR), and delay. Existing research primarily focuses on A2G channel modeling, often neglecting the impact of UAV heading. This paper addresses this gap by experimentally characterizing A2G and G2A channels in obstacle-free environments, considering distance and UAV heading as key parameters.

Existing literature presents various channel models for UAV communications, including A2G, G2G, and A2A. Some studies analyze link budgets, considering propagation losses and fading [6]. Others model A2G and G2A channels based on UAV altitude and Line-of-Sight (LoS) distance, analyzing path loss and fast-fading components [7]. Measurements of A2G channels for small UAVs at different altitudes and environmental conditions have also been reported, characterizing parameters like path loss, shadow fading, Doppler effect, Power Delay Profile (PDP), and Rician K-factor [2]. Empirical models for A2G and A2A propagation channels have been reviewed, classifying modeling approaches into deterministic, stochastic, and geometric-stochastic categories [8]. Other research explores the impact of receiver height on RF signal propagation [9], proposes new A2G channel models based on flight measurements [10], and investigates the variation of Angle of Arrival (AoA) and Angular Spread (AS) with altitude using LTE deployments [11]. Studies have also measured A2G channels at different frequencies and distances [12, 13], proposed statistical models for urban environments [14], and calculated path loss exponents and UDP throughput for air-ground-air (AGA) links [15]. However, a comprehensive characterization of RSSI and throughput, accounting for UAV headings and varying distances, has been lacking.

This paper employs an experimental approach to characterize A2G and G2A wireless channels using a testbed involving a UAV acting as an Access Point (AP) and a smartphone as a User Equipment (UE). The UAV is equipped with Wi-Fi and LTE communication modules, and omnidirectional antennas. The IEEE 802.11n (Wi-Fi 4) standard was used with MIMO 2x2, operating in channel 1 with a 20MHz bandwidth. An LTE base station (BS) provided LTE coverage. TCP traffic was generated from the UAV to the smartphone using iperf3, and the Wi-Fi MAC auto-rate mechanism (Minstrel) was used for link adaptation. Three experimental scenarios were conducted: Scenario A: The UAV was positioned 200m horizontally from the UE at 50m AGL. The UAV rotated in 45-degree increments to measure the Effective Radiation Pattern (ERP) of the antennas, considering the UAV's body's impact on signal reflection and obstruction. RSSI and downlink throughput were measured. Scenario B: The UAV moved away from and towards the UE (at 50m AGL), at 25m intervals maintaining a heading of 180º and then 0º. RSSI and downlink throughput were measured and compared with Friis and two-ray models. Scenario C: The UAV was positioned 1.42km from the LTE BS. Internet throughput was measured when the UE was directly connected to the LTE BS and when connected through the UAV via Wi-Fi. The system used an Alpha 800 UAV with a custom payload including an APU 4D4 board, a quad-core processor, 4GB RAM, OpenWRT 19.07.8 OS, and a 32GB SSD. A Xiaomi Mi 9T smartphone was used as the UE (single antenna). tcpdump was used on the UAV to measure RSSI.

Scenario A revealed a heterogeneous antenna radiation pattern due to the UAV's body obstructing the signal at certain headings. Summing the signals from both antennas improved the overall RSSI. High RSSI generally corresponded to high throughput, and vice-versa. Scenario B showed lower RSSI when the UAV moved away from the UE (180º heading) due to obstruction, and lower throughput during the return trip was attributed to limitations of the Minstrel auto-rate mechanism. Link asymmetry (10dB) and the fast ACK mechanism contributed to throughput variations. Scenario C demonstrated a 1.6x gain in throughput when connecting to the LTE BS through the UAV compared to a direct connection, highlighting the UAV's ability to overcome obstacles that blocked the direct LoS. Coverage and throughput were mapped, showing a coverage radius of up to 1500m with a dual-antenna UE and throughput varying from 10 Mbit/s to 80 Mbit/s in optimal directions.

The findings confirm the potential of UAVs to provide extended wireless connectivity, but also highlight the importance of considering real-world antenna radiation patterns and the impact of UAV orientation. The heterogeneous antenna radiation pattern necessitates considering UAV body effects in model development and system design. The performance limitations observed with the Minstrel algorithm suggest the need for more advanced link adaptation techniques. The superior performance when overcoming obstacles via the UAV demonstrates the potential for improved connectivity in challenging environments. These results provide valuable insights for designing efficient UAV communication systems.

This study successfully demonstrated internet connectivity between a UAV and UE up to 1500m, showcasing the potential of UAVs for wireless communication. The heterogeneous antenna radiation patterns and the impact of UAV heading and body obstruction were highlighted. Future work should focus on more sophisticated channel models incorporating UAV dynamics, improved link adaptation algorithms, and exploring different antenna configurations to optimize performance.

The experiments were conducted in an open, obstacle-free environment. The results might not be directly generalizable to more complex environments with obstacles or interference. The use of a single-antenna UE as a baseline may not fully represent the performance achievable with more advanced UE antenna configurations. The limited range of altitudes and distances in the experiments limits the scope of the model's applicability.