Temporal-Assisted Beamforming and Trajectory Prediction in Sensing-Enabled UAV Communications

TL;DR

This paper introduces a novel integrated sensing and communication framework for UAVs that uses temporal-assisted beamforming and trajectory prediction, significantly improving performance over traditional pilot-based systems.

Contribution

It develops a new ISAC framework with a temporal-assisted beamforming strategy using EKF, SCA, and IRM, tailored for UAV communication systems.

Findings

Achieves 156% improvement in multi-object scenarios

Achieves 136% improvement in single-object scenarios

Demonstrates effectiveness of temporal-assisted beamforming

Abstract

In the evolving landscape of high-speed communication, the shift from traditional pilot-based methods to a Sensing-Oriented Approach (SOA) is anticipated to gain momentum. This paper delves into the development of an innovative Integrated Sensing and Communication (ISAC) framework, specifically tailored for beamforming and trajectory prediction processes. Central to this research is the exploration of an Unmanned Aerial Vehicle (UAV)-enabled communication system, which seamlessly integrates ISAC technology. This integration underscores the synergistic interplay between sensing and communication capabilities. The proposed system initially deploys omnidirectional beams for the sensing-focused phase, subsequently transitioning to directional beams for precise object tracking. This process incorporates an Extended Kalman Filtering (EKF) methodology for the accurate estimation and prediction of object states. A novel frame structure is introduced, employing historical sensing data to optimize beamforming in real-time for subsequent time slots, a strategy we refer to as 'temporal-assisted' beamforming. To refine the temporal-assisted beamforming technique, we employ Successive Convex Approximation (SCA) in tandem with Iterative Rank Minimization (IRM), yielding high-quality suboptimal solutions. Comparative analysis with conventional pilot-based systems reveals that our approach yields a substantial improvement of 156\% in multi-object scenarios and 136\% in single-object scenarios.