[Paper Review] Mobile Edge Computing for Cellular-Connected UAV: Computation Offloading and Trajectory Optimization

Xiaowen Cao, Jie Xu+1 more|arXiv (Cornell University)|Mar 10, 2018

UAV Applications and Optimization8 references16 citations

TL;DR

This paper proposes a joint trajectory and computation offloading optimization framework for cellular-connected UAVs to minimize mission completion time by leveraging multiple ground base stations (GBSs) for parallel task execution. Using alternating optimization and successive convex approximation, the scheme achieves significant latency reduction—especially for high-computation tasks—by enabling dynamic UAV flight paths that maximize offloading efficiency.

ABSTRACT

This paper studies a new mobile edge computing (MEC) setup where an unmanned aerial vehicle (UAV) is served by cellular ground base stations (GBSs) for computation offloading. The UAV flies between a give pair of initial and final locations, during which it needs to accomplish certain computation tasks by offloading them to some selected GBSs along its trajectory for parallel execution. Under this setup, we aim to minimize the UAV's mission completion time by optimizing its trajectory jointly with the computation offloading scheduling, subject to the maximum speed constraint of the UAV, and the computation capacity constraints at GBSs. The joint UAV trajectory and computation offloading optimization problem is, however, non-convex and thus difficult to be solved optimally. To tackle this problem, we propose an efficient algorithm to obtain a high-quality suboptimal solution. Numerical results show that the proposed design significantly reduces the UAV's mission completion time, as compared to benchmark schemes.

Motivation & Objective

• Address the challenge of limited on-board computation resources in UAVs due to size, weight, and power (SWAP) constraints.
• Enable efficient computation offloading from UAVs to multiple cellular ground base stations (GBSs) with MEC capabilities.
• Minimize UAV mission completion time by jointly optimizing trajectory and offloading scheduling under speed, location, and GBS capacity constraints.
• Overcome the non-convexity of the joint optimization problem through a suboptimal but efficient algorithmic design.

Proposed method

• Formulate a non-convex optimization problem to minimize UAV mission completion time via joint trajectory and computation offloading scheduling.
• Apply alternating optimization to decouple trajectory and offloading variables, solving them iteratively.
• Use successive convex approximation (SCA) to iteratively approximate the non-convex constraints into convex ones for tractable solution.
• Design an initial heuristic trajectory (straight flight) and refine it using bisection search over time duration to ensure feasibility.
• Optimize time allocation and hovering durations in the successive-hover-and-fly benchmark for performance comparison.
• Implement a sampling-based trajectory projection (every 1s) to visualize and analyze optimized flight paths under varying task loads.

Experimental results

Research questions

• RQ1 How can UAV trajectory and computation offloading be jointly optimized to minimize mission completion time in a cellular-connected MEC setup?
• RQ2 What is the impact of task size on the optimal UAV flight strategy and offloading efficiency?
• RQ3 How does the proposed joint optimization compare to benchmark schemes like straight flight and successive hover-and-fly in terms of latency reduction?
• RQ4 To what extent can dynamic trajectory adaptation improve offloading performance when computation capacity is constrained?

Key findings

• For small task sizes (e.g., L = 100 Mbits), the straight-flight scheme achieves the same mission completion time as the proposed design, as flight time dominates.
• When task size increases (e.g., L = 200 Mbits), the UAV deviates from a straight path to fly closer to GBSs 1, 4, and 5 to exploit better channel conditions for offloading.
• For large tasks (L = 500 Mbits), the UAV hovers above and flies back and forth between GBSs 4 and 5, indicating computation time becomes the main bottleneck.
• The proposed design outperforms both benchmark schemes across all task sizes, with the performance gap widening as task size increases.
• The successive-hover-and-fly scheme performs nearly as well as the proposed design at L = 300 Mbits but lags significantly at higher loads due to fixed visiting order and longer flight paths.
• Numerical results confirm a significant reduction in mission completion time using the proposed algorithm, especially under high-computation workloads.