AERPAW Research Platform

• AERPAW platform is a NSF-funded infrastructure integrating UAVs, SDRs, and ground sensors to enable realistic wireless research experiments.
• It employs digital twin emulation alongside real-world testing to assess advanced algorithms under dynamic channel and mobility constraints.
• Quantitative evaluations show that novel HGAD methods outperform traditional greedy approaches by enhancing throughput and reducing handovers.

The AERPAW (Aerial Experimentation and Research Platform for Advanced Wireless) platform is a National Science Foundation-funded infrastructure designed for research in realistic wireless networking environments, integrating multi-rotor unmanned aerial vehicles (UAVs), programmable radios, and ground sensor nodes. It supports both digital twin (DT) emulation and real-world (RW) experimentation, enabling rigorous validation of advanced wireless algorithms that are aware of real propagation phenomena, UAV mobility constraints, and mission timing limitations. The platform plays a prominent role in bridging the gap between idealized network simulations and the operational complexities of field deployments, notably for research on UAV-enabled data-mule strategies under time and wireless channel constraints (Hossen et al., 9 Jan 2026).

1. Platform Architecture and Capabilities

AERPAW supports coordinated experimentation with UAVs equipped with SDRs (Software Defined Radios) such as the USRP B205mini, interfaced with multiple ground sensors (e.g., USRP B210) over sub-6 GHz links. The platform is geographically situated to expose channels to realistic outdoor fading, terrain shadowing, and obstructions. Flight operations within AERPAW are executed under strict geofencing rules, with the UAV’s real-time 2-D position $r_t$ constrained in space $\mathcal{G}$ and mobility limited by $‖r_{t+1} - r_t‖ \leq v_{\max} Δt$ per mission slot.

Experimental capabilities include logging per-sensor instantaneous signal-to-noise ratio (SNR), mapping SNR to achievable rates using modulation and coding scheme (MCS) tables, and enabling fine-grained trajectory control and per-second SNR/rate trace collection. The digital twin mode of AERPAW incorporates realistic channel modeling—via actual emulation software—using terrain data and fading/replay of traces from flights for repeatable tests.

2. Problem Setting: UAV Data Mule Missions

UAV data-mule operations at AERPAW typically involve a multi-rotor UAV tasked with downloading buffered data from a geographically dispersed set of $N$ ground sensors within a hard mission time $T$. Each sensor maintains a data buffer $Q_i$ to be emptied before mission completion. The wireless channel between UAV and each sensor $i$ experiences large, often unpredictable SNR fluctuations as a function of $r_t$, primarily induced by UAV mobility, environmental shadowing, and multipath fading. Standard signal-strength-only (greedy) association policies often result in excessive handovers, buffer starvation for large sensors, and underutilization of peak channel intervals.

3. Mathematical Model for Data-Driven Evaluation

Experiments at AERPAW are formulated under the following mathematical model. Let $S = \{1, \dots, N\}$ index the sensor set, initial sensor bufffers $Q_i(0) = Q_i > 0$, and the time horizon be slotted $t = 0, \dots, T-1$ of duration $Δt$. At most one sensor may be served per slot: $\sum_{i \in S} x_{i,t} \leq 1$, $x_{i,t} \in \{0,1\}$. SNR in dB for sensor $i$ at position $r$ is

$$\mathsf{SNR}_i^{\mathsf{dB}}(r) = P_{tx}^{\mathrm{dBm}} + G_{tx}^{\mathrm{dBi}} - PL(r)^{\mathrm{dB}} + G_{rx}^{\mathrm{dBi}} - N_0^{\mathrm{dBm}}$$

and rate is mapped as $R_i(r) = f(\mathsf{SNR}_i(r))$. Per-slot downloaded bits $0 \leq y_{i,t} \leq R_i(r_t) x_{i,t} Δt$, with buffer evolution $Q_i(t+1) = \max\{Q_i(t) - y_{i,t},0\}$. The objective is to maximize total downloaded bits subject to mobility and per-sensor buffer/download constraints.

4. Evaluation Methodology and Experimental Workflow

AERPAW supports three primary experimental modalities for UAV data-mule algorithm evaluation:

• Simulation: Uses free-space path loss and ideal SNR-to-rate mapping (MCS), four sensors with fixed initial buffer sizes (e.g., [500, 800, 700, 1000] Mbits), mission duration $T=500$ s, and controlled UAV trajectory.
• Digital Twin (DT): Implements terrain-aware channel emulation, terrain/fading-aware path loss, and path/trace replay with actual AERPAW channel software, matching the real-world deployment.
• Real-World (RW) Testbed: Employs a hardware-in-the-loop UAV (e.g., USRP B205mini) over actual flight paths, ground nodes (USRP B210), and empirically measured SNRs, with data rates mapped in real time via MCS.

Representative mission configurations include "Flight 1" (dense sensor region, $T=360$ s, buffers=[500,800,700,1000] Mbit) and "Flight 2" (wide-area sweep, $T=1100$ s, buffers=[1500,1300,1100,200] Mbit).

5. Algorithmic Baselines and the HGAD Approach

The AERPAW platform has enabled rigorous comparative performance studies of association and scheduling algorithms. The primary reference benchmark is the traditional Greedy approach, where the UAV always selects the sensor with the strongest instantaneous SNR, disregarding buffer sizes and temporal channel correlations. This method is mathematically described as

$$i_{\text{Greedy}}(t) = \arg\max_{i: Q_i(t)>0} \mathsf{SNR}_i(r_t)$$

A novel algorithm, Hover-based Greedy Adaptive Download (HGAD), was introduced and evaluated on AERPAW to address known deficiencies of the Greedy baseline. HGAD selects the sensor with the highest rate (not merely SNR), prefers to hover over sensors when the rate is near its historical peak, and schedules hover duration as

$$T^{\text{hover}}_i = \min\{Q_i^{\text{rem}}/R_i^{\max}, T_i^{\max}\}$$

where $Q_i^{\text{rem}}$ is remaining buffer, $R_i^{\max}$ the sensor maximum rate, and $T_i^{\max}$ a mission-imposed hover time limit. The rate-based selection and buffer-aware hovering are designed to maximize throughput and ensure completion of large buffers.

6. Quantitative Results and Performance Metrics

AERPAW's data traces substantiate the superiority of HGAD over the Greedy baseline across all modalities. Quantitative comparisons, as shown in the following table, clarify mission efficiency improvements:

Scenario	Trajectory	Greedy (Mbit)	HGAD (Mbit)	Mission T (s)
Digital Twin	Fixed	1,233	1,930 (+57%)	500
Simulation	Fixed	2,218	2,518 (+14%)	500
Simulation	Autonomous	2,223	2,864 (+29%)	500
RW Flight 1	Fixed	563	787 (+40%)	360
RW Flight 2	Fixed	2,002	3,944 (+97%)	1100

Per-sensor buffer fulfillment for RW Flight 2 further reveals that under Greedy, large-buffer sensors (e.g., LW2, LW3, LW4) may not be fully serviced (e.g., Greedy achieves only $\sim$800/1,300 Mbit for LW2, only $\sim$200/1,100 Mbit for LW3, and just 2/200 Mbit for LW4), whereas HGAD consistently achieves full buffer downloads.

HGAD also reduces handover rates by approximately 70% and increases hover ratio (stationary time at sensor), e.g., from ~10% to ~35%.

7. Impact and Research Significance

Experiments on AERPAW demonstrate that integrating SNR-to-rate awareness, buffer tracking, and real-time hover logic are essential for robust UAV data-mule performance in realistic mission scenarios. The platform validates that naive signal-strength-based switching frequently leads to excess motion overhead and buffer starvation, while the HGAD approach consistently yields 40–100% higher throughput, achieves fair per-sensor buffer depletion, and delivers stable, high-rate data intervals under practical deployment constraints.

AERPAW's combination of digital twin and real-world emulation supports reproducible cross-validation of theoretical algorithms, enables assessment on both synthetic and empirically observed channel dynamics, and provides a benchmark for developing resilient field-grade algorithms for tactical and emergency UAV deployments (Hossen et al., 9 Jan 2026).