Edge Computing Aware NOMA for 5G Networks

Abbas Kiani, Nirwan Ansari

TL;DR

This paper proposes a novel approach to optimize energy consumption in 5G mobile edge computing (MEC) networks by jointly considering Non-Orthogonal Multiple Access (NOMA) and edge computing resource allocation. The authors develop a framework that minimizes MEC users' uplink energy consumption by intelligently clustering users, allocating communication and computing resources, and controlling transmit powers, offering a more efficient solution than traditional methods.

What Problem Does This Paper Address?

The rapid growth of the Internet of Things (IoT) and the increasing demand for sophisticated mobile applications are pushing 5G wireless networks to their limits. Two key technologies emerging to meet these demands are Non-Orthogonal Multiple Access (NOMA) and Mobile Edge Computing (MEC). NOMA promises to significantly increase network capacity by allowing multiple users to share the same radio resources, while MEC aims to reduce latency and improve Quality of Service (QoS) by bringing computing capabilities closer to users.

However, effectively integrating these technologies presents a challenge. While NOMA can boost capacity and MEC can reduce latency, their combined potential for optimizing overall system performance, particularly in terms of energy efficiency for MEC users, has not been fully explored. Specifically, the paper identifies a gap in research concerning the joint optimization of communication and computing resources within a NOMA framework for MEC, considering the complexities of intra-cell interference and user deadlines. This paper aims to bridge this gap by proposing an "edge computing aware NOMA" technique.

Methods

The core of this paper lies in its proposed optimization framework and a two-phase approach to solve it.

  1. Optimization Framework (Problem P1): The authors formulate a Mixed Integer Non-Linear Programming (MINLP) problem to minimize the total uplink energy consumption of MEC users. This problem jointly optimizes:
  2. User Clustering: Grouping users into NOMA clusters.
  3. Resource Allocation: Assigning frequency Resource Blocks (RBs) and computing RBs to users.
  4. Transmit Power Control: Determining the power levels for users. The framework incorporates constraints such as user deadlines, minimum data rate requirements, maximum transmit power, and the total available computing and communication resources.

  5. Two-Phase Solution Approach: Recognizing that solving the MINLP problem directly is computationally intractable, the authors propose a two-phase approach:

  6. Heuristic Algorithm for User Clustering and RB Allocation: This algorithm efficiently determines the binary and integer variables (user clustering, RB assignments) in three separated phases: user clustering based on average channel conditions, computing RB allocation considering deadlines, and frequency RB allocation to meet data rate requirements.
  7. Convex Optimization for Power Control: Once the clustering and RB allocations are fixed, the problem is reduced to optimizing transmit powers. The authors reformulate the problem (P2 and P3) into a convex optimization problem, which can be efficiently solved using interior-point methods. This involves a change of variables and approximations for high Signal-to-Interference-plus-Noise Ratio (SINR) regimes to ensure convexity.

Datasets

Not discussed in the paper. The evaluation is performed using simulations with parameters described in Section IV.

Key Insights and Ideas

  • Synergy of NOMA and MEC for Energy Efficiency: The paper demonstrates that by intelligently combining NOMA's capacity gains with MEC's proximity benefits, significant reductions in MEC users' uplink energy consumption can be achieved. This is a crucial insight for battery-constrained IoT devices.
  • Joint Optimization is Key: The research highlights that optimizing communication and computing resources in isolation is suboptimal. A joint optimization approach, considering the interplay between these resources, leads to better overall system performance.
  • Deadline-Aware Resource Allocation: Incorporating user deadlines into the resource allocation process is critical for real-time MEC applications. The proposed heuristic algorithm effectively balances resource allocation with these time constraints.
  • Heuristic Approach for Tractability: The development of an efficient heuristic algorithm is a practical contribution, enabling near-optimal solutions in a reasonable time, which is essential for real-world deployment where optimal solutions are often computationally prohibitive.
  • Impact of Umax on Performance: The study shows that increasing the maximum number of users per frequency RB (Umax) can improve spectral efficiency, but also leads to increased intra-cell interference, creating a trade-off that needs careful management.

Key Results

  • Energy Consumption Reduction: The proposed heuristic algorithm achieves total energy consumption close to the optimal solution, significantly outperforming approaches that do not consider joint optimization. For instance, Figure 1 shows the heuristic approach yielding results very close to the optimal approach.
  • Improved Spectral Efficiency: Figure 3 demonstrates that spectral efficiency increases with the number of users and average input data, indicating the effectiveness of the NOMA scheme in utilizing resources.
  • Fairness in Computing Resource Allocation: Figure 5 illustrates that finer granularity in computing RBs (smaller capacity per RB) leads to fairer allocation of computing time among users, which is important for user experience.
  • Impact of Computing Capacity and RB Division: Figure 4 shows that dividing total computing capacity into smaller RBs can improve energy efficiency, especially for a higher number of users, due to more equitable resource distribution.

Limitations and Future Work

  • Contiguous RB Constraint: The paper acknowledges that the proposed frequency RB allocation does not consider the contiguity constraint (RBs allocated to a user must be contiguous). They suggest that this can be addressed by adopting RB grouping algorithms, but note that this extension is NP-hard.
  • Channel State Information (CSI): The current model assumes that CSI is known to the base station (BS). Future work could explore scenarios with partial or imperfect CSI.
  • Scalability: While the heuristic approach improves tractability, the complexity of the power control optimization problem can still increase with the number of users per NOMA cluster. Future research could focus on further reducing this complexity.
  • Dynamic Environments: The current work assumes a relatively static environment. Investigating the performance in highly dynamic environments with rapidly changing channel conditions and user mobility would be a valuable extension.

Summary updated

BibTeX
@article{kiani2018edge,
  author = {Abbas Kiani and Nirwan Ansari},
  title = {{Edge Computing Aware NOMA for 5G Networks}},
  journal = {IEEE Internet of Things Journal},
  year = {2018},
  doi = {10.1109/jiot.2018.2796542},
}
Figure