Source-linked AI summary
UAV-Enabled Mobile Edge Computing: Offloading Optimization and Trajectory Design
Fuhui Zhou, Yongpeng Wu, Haijian Sun, Zheng Chu
TL;DR
Mobile users’ latency-sensitive applications are constrained by limited computation capability and battery lifetime, while MEC and wireless power transfer face propagation-related limitations. The paper studies UAV-enabled wireless powered MEC, jointly formulates computation and trajectory optimization as a non-convex power-minimization problem, and solves it with an alternative sequential-convex-optimization algorithm. Simulations show that the proposed design outperforms benchmark schemes and converges in only several iterations.
Problem
Mobile applications require substantial computation and low latency, but mobile devices have limited computation capability and battery lifetime, while wireless power transfer suffers propagation loss.
Method
The paper jointly optimizes offloading bits, users’ and UAV CPU frequencies, and UAV trajectory in a power-minimization problem using an alternative algorithm based on sequential convex optimization.
Results
Simulation results show that the proposed design outperforms benchmark schemes and that its algorithm converges in only several iterations.
Takeaways & Limitations
UAV-enabled wireless powered MEC can jointly provide computation offloading and sustainable operation opportunities for multiple ground users within the studied system model.
Abstract
from arXiv · showhide
With the emergence of diverse mobile applications (such as augmented reality), the quality of experience of mobile users is greatly limited by their computation capacity and finite battery lifetime. Mobile edge computing (MEC) and wireless power transfer are promising to address this issue. However, these two techniques are susceptible to propagation delay and loss. Motivated by the chance of short-distance line-of-sight achieved by leveraging unmanned aerial vehicle (UAV) communications, an UAV-enabled wireless powered MEC system is studied. A power minimization problem is formulated subject to the constraints on the number of the computation bits and energy harvesting causality. The problem is non-convex and challenging to tackle. An alternative optimization algorithm is proposed based on sequential convex optimization. Simulation results show that our proposed design is superior to other benchmark schemes and the proposed algorithm is efficient in terms of the convergence.
I. INTRODUCTION
The paper studies UAV-enabled wireless powered MEC to address mobile users’ limited computation capability and battery lifetime. It jointly optimizes computation and UAV trajectory through a power-minimization formulation solved with sequential convex optimization.
- Mobile applications are latency-sensitive and difficult for devices with limited battery and computation capability to execute.
- WPT provides controllable energy harvesting but suffers from propagation loss, motivating UAV-enabled wireless power transfer.
- The system lets a UAV transmit energy to multiple ground users that use harvested energy for local computing and task offloading.
- The paper formulates a non-convex power-minimization problem jointly optimizing offloading bits, users’ and UAV CPU frequencies, and UAV trajectory.
- An alternative optimization algorithm based on sequential convex optimization is proposed, and the resource-allocation scheme outperforms benchmark schemes.
II. SYSTEM MODEL
The system uses a fixed-altitude UAV with an MEC server to power multiple ground users, who simultaneously harvest energy, compute locally, and offload tasks. Channel, harvesting, computation, and propulsion models define the variables and energy costs used for optimization.
- The UAV carries an MEC server, transmits energy to K ground users, and supports partial computation offloading.
- Users simultaneously perform energy harvesting, local computing, and computation offloading during discretized time slots.
- The UAV flies at fixed altitude over a finite horizon divided into N equal-time slots, with users fixed on the ground.
- A line-of-sight channel model determines channel power gain from the UAV-user distance, and a linear energy-harvesting model determines harvested energy.
- TDMA divides each slot among users for offloading, while local-computation and UAV offloading energies depend on CPU frequencies and computation models.
- The propulsion-energy model depends only on UAV velocity; a more general velocity-and-acceleration model is left for future work.
A. The Energy Minimization Problem Formulation
The paper formulates a UAV energy minimization problem that jointly allocates computation and UAV movement while enforcing computation, energy-causality, timing, speed, and endpoint constraints.
- The objective jointly optimizes users’ offloading bits, users’ and UAV CPU frequencies, and the UAV trajectory to minimize energy consumption.
- Users do not offload in the final slot, the UAV does not compute in the first slot, and the UAV follows speed and initial/final-location constraints.
- The formulation requires each user to complete its total computation-bit demand while respecting causal harvested-energy availability.Constraint C2 limits local-computing and offloading energy to harvested energy.
- The UAV cannot compute more bits in a slot than users have offloaded before the preceding slot.
- The resulting problem is non-convex because the UAV trajectory, offloading bits, and CPU frequencies are coupled.
B. Computation Offloading And CPU Frequency Optimization
For a fixed UAV trajectory, the computation-offloading and CPU-frequency subproblem becomes convex and can be solved through Lagrange duality, yielding structural insights about allocation.
- Fixing the UAV trajectory makes the offloading and CPU-frequency subproblem convex.
- The convex subproblem can be solved using Lagrange duality to obtain optimal offloading bits and CPU frequencies.
- The solution uses dual variables associated with total computation, energy causality, slotwise UAV computing, and complete offloading constraints.
- The UAV CPU frequency increases across time slots, so users should allocate more energy to local computation to reduce offloading and UAV energy.
- Improved UAV-user channel conditions increase users’ offloading bits, corresponding to shorter UAV-user distances.
C. Trajectory Optimization
The trajectory subproblem is handled by alternating between convex computation optimization and sequential-convex-approximation trajectory updates, producing an algorithm that converges in only several iterations.
- For fixed offloading bits and CPU frequencies, trajectory optimization is formulated as a separate problem.
- The trajectory problem is non-convex because of the energy-causality constraint.
- Sequential convex approximation replaces the non-convex trajectory step with iteratively solved approximate problems.The resulting solutions satisfy the Karush-Kuhn-Tucker conditions of the trajectory subproblem.
- Each approximate trajectory problem retains the UAV speed and initial/final-location constraints.
- The alternative optimization algorithm combines the computation and trajectory subproblems, with its objective value denoted by Eᵤ.
IV. SIMULATION RESULTS
The proposed design is evaluated against two benchmark trajectories, with results covering UAV paths, energy consumption, and algorithm convergence. The proposed scheme achieves the lowest reported UAV energy consumption and converges in only several iterations.
- Experimental setup: The proposed design is compared with a constant-speed straight-flight scheme and a semicircular trajectory scheme.Scheme 1 flies straight from the initial to final position, while Scheme 2 follows a semicircle whose diameter is the distance between those positions.
- Trajectory comparison: Under the proposed trajectory, the UAV first tends toward Users 2 and 3, then flies faster toward the final position.Users 2 and 3 require more energy because they have larger numbers of computation bits to be offloaded.
- Energy consumption: The proposed scheme consumes the smallest total UAV energy among the evaluated schemes.The comparison jointly optimizes offloading bits, user and UAV CPU frequencies, and UAV trajectory.
- Energy consumption: Total UAV energy consumption decreases as the UAV flying time increases for every evaluated scheme.The paper attributes this trend to lower flying speeds and UAV CPU frequencies at longer flying times.
- Convergence: Algorithm 1 requires only several iterations to converge.Figure 4 evaluates the convergence efficiency of the proposed alternative algorithm.
V. CONCLUSION
The paper studies UAV-enabled wireless powered MEC in which joint resource and trajectory optimization minimizes UAV energy consumption. Its design outperforms benchmark schemes and converges in only several iterations.
- Conclusion: The system jointly optimizes offloading bits, user and UAV CPU frequencies, and UAV trajectory to minimize UAV energy consumption.The UAV provides multiple ground users with computation offloading and sustainable operation opportunities.
- Conclusion: The proposed alternative algorithm uses sequential convex approximation techniques.The conclusion reports this algorithm as the solution approach for the optimization problem.
- Conclusion: The proposed design outperforms other benchmark schemes and the algorithm requires only several iterations to converge.These outcomes are reported from the simulation results.
APPENDIX A PROOF OF THEOREM 1
The appendix derives conditions for the optimization variables by differentiating the Lagrangian of problem P2. Setting these derivatives to zero yields the stated equations and completes the proof.
- Lagrangian derivation: The proof uses the Lagrangian of P2, with dual variables associated with constraint C5 and the optimization and dual-variable set Ξ1.The dual variables ρk and ϑ are nonnegative.
- Lagrangian derivation: Derivatives of the Lagrangian with respect to lk[n] and fk[n] are derived for the proof.The appendix introduces these derivatives after defining the relevant Lagrangian quantities.
- Proof completion: Setting the relevant derivatives to zero yields equations (8a), (8b), and (8c), completing the proof of Theorem 1.Equation (8b) follows from equation (15).