Abstract
Migrating computational intensive tasks from mobile devices to more resourceful cloud servers is a promising technique to increase the computational capacity of mobile devices while saving their battery energy. In this paper, we consider an MIMO multicell system where multiple mobile users (MUs) ask for computation offloading to a common cloud server. We formulate the offloading problem as the joint optimization of the radio resources—the transmit precoding matrices of the MUs—and the computational resources—the CPU cycles/second assigned by the cloud to each MU—in order to minimize the overall users' energy consumption, while meeting latency constraints. The resulting optimization problem is nonconvex. Nevertheless, in the single-user case, we are able to compute the global optimal solution in closed form. In the more challenging multiuser scenario, we propose an iterative algorithm, based on a novel successive convex approximation technique, converging to a local optimal solution of the original nonconvex problem. We then show that the proposed algorithmic framework naturally leads to a distributed and parallel implementation across the radio access points, requiring only a limited coordination/signaling with the cloud. Numerical results show that the proposed schemes outperform disjoint optimization algorithms.
Introduction
Mobile devices are currently caught in a "resource pincer": modern applications like augmented reality, high-end gaming, and AI-driven assistants are becoming increasingly computationally intensive, yet battery technology improvements remain incremental. Mobile-Edge Computing (MEC) offers a solution by offloading heavy tasks to nearby servers.
However, offloading introduces its own costs in terms of energy (for data transmission) and latency. In dense multi-cell MIMO environments, this is further complicated by inter-cell interference. If one user increases transmit power to offload faster, they may interfere with neighbors, creating a cycle of increased energy consumption. This paper addresses these challenges by treating radio and computational resources as a single coupled optimization problem rather than managing them disjointly.
Methodology
The authors formulate the problem as a sum-energy minimization subject to transmit power and latency constraints. The latency constraint acts as the "glue," representing the sum of transmission time (radio) and execution time (computation).
Single-User Case
For a single user, the authors demonstrate that the non-convex energy minimization problem can be transformed into a convex one. They derive a closed-form global optimal solution resembling a modified "water-filling" technique, where the resource allocation is determined by both radio channel conditions and CPU parameters. To handle the specific computation of the optimal water level (represented by $\alpha$), they developed an efficient iterative routine.
Multi-User Case
The multi-user scenario is significantly more complex due to non-convex interference terms. To solve this, the authors propose an Inner Successive Convex Approximation (SCA) algorithm. This method solves a sequence of simpler, convex approximations that are guaranteed to converge to a local optimum of the original problem.
Distributed Implementation
To ensure practical feasibility, the authors propose distributed versions of the algorithm. By utilizing dual decomposition or slack variables to decouple interference terms, base stations can make local decisions with minimal signaling to the central cloud controller.
Experimental Setup
The framework was validated through numerical simulations with the following parameters:
- Network Setup: 2 cells with 4 users each.
- Hardware: MIMO configurations ranging from $2 \times 2$ to $2 \times 8$ antennas.
- Channel Model: Rayleigh fading channels, with results averaged over 100 independent realizations.
- Benchmarks: The proposed joint optimization was compared against Disjoint Resource Allocation (DRA), where CPU cycles are assigned proportionally to workload without considering the radio environment.
Key Results and Insights
- The Coupling Advantage: The optimal transmit strategy is deeply tied to the CPU budget. If the cloud has surplus CPU cycles, the mobile device can transmit slower to save energy, as the cloud can "make up the time" through faster processing.
- Energy Efficiency: Joint optimization significantly outperforms DRA, particularly when the "computation-to-data" ratio is low (i.e., data-heavy tasks).
- Convergence and Robustness: The SCA algorithm typically reaches a stable solution in fewer than 10 iterations and proved insensitive to initialization, consistently converging to the same result from various starting points.
- MIMO Scaling: Increasing the number of receive antennas at the base station (e.g., from $2 \times 2$ to $2 \times 8$) drastically reduces the transmit energy required by mobile devices by improving signal reception in the presence of interference.
Limitations and Future Work
While the framework is robust, it currently assumes perfect Channel State Information (CSI), which may be difficult to maintain in high-mobility scenarios. Additionally, the current model focuses on a static snapshot in time. Future research could extend this to dynamic queueing models where tasks arrive at random intervals. Finally, while the current objective minimizes total system energy, future iterations could incorporate explicit fairness constraints to ensure users at the cell edge are not disadvantaged.
Summary updated