Abstract
The rapid development of the Internet of Things (IoT) has led to a surge in delay-sensitive and computation-intensive applications. Due to the inherent latency of cloud computing and the limited resources of end devices, Mobile Edge Computing (MEC) has emerged as a critical solution. This paper investigates energy-efficient computation offloading in 5G heterogeneous networks, where devices can access multiple edge servers. We propose an iterative framework to jointly optimize transmission power and offloading decisions, achieving near-optimal energy efficiency while meeting strict latency requirements.
Introduction
Modern IoT applications, such as augmented reality and real-time sensing, demand significant computational power and low latency. Traditional cloud-based offloading often fails to meet these requirements due to propagation delays. MEC addresses this by deploying servers at the network edge. However, the transition to 5G introduces heterogeneous architectures where devices are within range of multiple access points (APs) and base stations (BSs). This creates a complex optimization problem: how to distribute tasks across multiple heterogeneous servers to minimize total energy consumption without violating latency constraints.
System Architecture
The proposed framework operates across three layers: the management layer, the edge layer, and the device layer.
The Device Layer consists of various IoT devices (e.g., wearables, cameras) with computation-intensive tasks. The Edge Layer contains heterogeneous servers, such as macro base stations and small-cell access points, each with different computing capacities and channel conditions. The Management Layer coordinates the network information and control signaling to optimize offloading decisions.
Methodology
The authors formulate an Energy Minimization (EM) problem, categorized as a Mixed-Integer Nonlinear Program (MINLP). To handle its complexity, the problem is decomposed into two subproblems:
- Transmission Power Allocation (TPA): Optimizing the power used by devices to transmit data to their assigned servers, ensuring deadlines are met with minimal energy.
- Computation Offloading Decision (COD): Determining the optimal server assignment for each device.
An iterative solution framework is proposed, consisting of an Initialization Algorithm (IA) based on channel gain, followed by a Device Reassignment Algorithm (DRA) and a Device-Server Pair Exchange Algorithm (DSPEA) to refine the assignments and escape local optima.
Key Insights and Results
Simulation results demonstrate that the proposed iterative framework is highly effective, achieving energy consumption levels within $3 \times 10^{-4}$ of the global optimum.
The Benefit of Heterogeneity
A key finding is that distributing computing resources across multiple edge servers is more energy-efficient than a centralized approach. By spreading resources, devices can connect to closer nodes, significantly reducing transmission power. The study suggests an optimal "sweet spot" for resource distribution; for example, an equal split of resources between an AP and a BS often yields the best performance in uniform device distributions.
Scalability and Convergence
The algorithm exhibits a computational complexity of $O(M^2)$, where $M$ is the number of devices. In practice, it converges rapidly—typically within 2 to 4 iterations—making it suitable for dynamic network environments.
Conclusion
This research provides a robust framework for managing computation offloading in 5G heterogeneous networks. By intelligently leveraging multiple edge resources, the proposed method significantly reduces the energy footprint of IoT devices while maintaining the millisecond-level latency required for next-generation applications. Future work will explore hardware heterogeneity (e.g., GPUs) and high-mobility scenarios.
Summary updated