Sensing-Assisted Predictive Beamforming for UAV-Enabled Ocean Monitoring Networks

Technical Summary: Sensing-Assisted Predictive Beamforming for UAV-Enabled Ocean Monitoring Networks

Problem Statement
Ocean monitoring networks rely on surface buoys to collect environmental data, but these buoys face significant challenges in data transmission due to their mobility and the lack of stable infrastructure backhaul. Unmanned Aerial Vehicles (UAVs) offer a flexible solution for data collection via line-of-sight (LoS) links. However, the continuous motion of buoys induced by sea waves and currents causes rapid variations in relative range and angles. This mobility is particularly detrimental when UAVs employ large antenna arrays and highly directional beams, as even minor pointing mismatches lead to severe beamforming gain loss and link outages. Furthermore, sea clutter degrades the accuracy of sensing echoes, complicating the acquisition of buoy state information. Existing works often assume static sea-surface nodes or utilize motion models (e.g., constant velocity) that fail to capture the wave-driven oscillations and current-induced drift characteristic of ocean buoys. Consequently, there is a lack of a unified framework that jointly addresses wave-driven buoy dynamics, residual sea clutter, and predictive beam alignment for UAV-enabled maritime data collection.

Methodology
The paper proposes a sensing-assisted predictive beamforming framework integrated into a frame-based UAV mission workflow. The core methodology involves the following components:

1. System Model and Workflow: The mission is divided into radio frames. Within each operation block (two consecutive frames), the UAV transmits Integrated Sensing and Communication (ISAC) signals. These signals serve a dual purpose: acquiring buoy echoes for state estimation and supporting subsequent uplink communication. The workflow involves predicting buoy states from the previous frame, transmitting ISAC signals, refining state estimates using received echoes via an Extended Kalman Filter (EKF), and generating proactive beam predictions for the next frame.
2. Physics-Grounded Motion Modeling: To characterize short-horizon buoy motion, the authors develop a correlated-acceleration state-space model. This model combines a Singer process to represent wave-driven excitation (modeled as a first-order Gaussian Markov process) with a slowly varying current-drift term. The Singer correlation time is explicitly linked to the dominant wave time scale, making the model suitable for maritime environments.
3. Clutter-Aware Sensing and PCRB Derivation: The signal model explicitly accounts for residual sea clutter, which is modeled as a low-power interference term sharing the target's Doppler bin after suppression. The authors derive the posterior Fisher Information Matrix (FIM) and the corresponding Posterior Cramér–Rao Bound (PCRB) for the nonlinear sensing measurement model. The predicted horizontal-position PCRB is adopted as the sensing metric to quantify the impact of sea-surface dynamics and clutter on predictive beam alignment.
4. Optimization Formulation: A "worst-buoy" design problem is formulated to jointly optimize sensing power allocation and UAV position. The objective is to minimize the maximum PCRB among all buoys (maximizing the worst-case sensing accuracy) while satisfying uplink rate, UAV power, and mobility constraints.
5. Algorithmic Solution: The resulting optimization problem is non-convex due to the coupling of variables in the FIM. The authors employ a Schur-complement reformulation to handle the PCRB objective and a "lagged successive convex approximation" (SCA) strategy. In this approach, complex matrix terms in the FIM are updated in a lagged manner (using values from the previous iteration), while scalar weights are convexified via first-order Taylor expansion. This converts the subproblem into a convex conic program with tractable complexity.

Key Contributions
The paper claims the following contributions:

• Framework Development: Establishment of a clutter-aware, sensing-assisted predictive beamforming framework that tightly integrates target sensing, EKF-based state refinement, proactive beam prediction, and uplink collection under sea-surface dynamics.
• Motion Modeling: Development of a tractable buoy-motion model that jointly captures wave-induced oscillations and current-induced drift, linking the Singer correlation time to the dominant wave time scale.
• Theoretical Analysis: Derivation of the posterior FIM and PCRB for the nonlinear sensing model in the presence of residual sea clutter, providing a principled metric for predictive beam alignment design.
• Optimization Algorithm: Formulation of a joint sensing-power and UAV-position design problem and its solution via a lagged-SCA algorithm, which converts the problem into a convex conic program with complexity scaling linearly with the number of buoys in the decision variables and cubically in the total problem size.
• Performance Validation: Demonstration through extensive simulations that the proposed scheme maintains robust performance under denser buoy deployments, harsher sea conditions, and higher noise levels, outperforming several baseline designs.

Simulation Results
Numerical results indicate that the proposed design achieves the most balanced sensing–communication performance among the considered benchmarks.

• Robustness: The scheme remains effective as buoy density increases, process/measurement noise intensifies, and sea conditions become harsher (higher sea state indices).
• Necessity of Sensing: A comparison with a "Communication-Only" benchmark (which relies on open-loop state propagation without sensing updates) reveals a pronounced Root Mean Square Error (RMSE) degradation in the benchmark. This confirms that sensing-assisted state refinement is essential for accurate predictive beamforming in dynamic maritime environments.
• Complexity vs. Performance: When compared to a "Full First-Order Taylor Expansion" (Full-FOTE) method, the proposed lagged-SCA approach incurs only a moderate performance loss (e.g., a slight reduction in uplink rate) but offers a significantly lower computational burden. Specifically, the Full-FOTE method requires substantially more optimization time as the number of buoys increases, highlighting the practical advantage of the proposed method for online deployment.

Significance and Claims
The paper asserts that its primary significance lies in bridging the gap between maritime radar modeling (clutter and wave dynamics) and communication beam management. By explicitly modeling the physics of buoy motion and sea clutter, the proposed framework enables robust UAV-buoy communication where traditional static or simple kinematic models fail. The authors claim that their approach provides a practical trade-off between performance and computational complexity, making it suitable for real-time online deployment in offshore monitoring networks. The work emphasizes that without sensing-assisted refinement, predictive beamforming in such dynamic environments is unreliable, and that the proposed lagged-SCA strategy offers a computationally efficient path to achieving robust connectivity. Future work is noted to extend this centralized single-UAV design to multi-UAV cooperative scenarios.