[Paper Review] Joint Computation and Communication Resource Optimization for Beyond Diagonal UAV-IRS Empowered MEC Networks

Asad Mahmood, Thang X. Vu+3 more|arXiv (Cornell University)|Nov 13, 2023

IoT and Edge/Fog Computing4 citations

TL;DR

This paper proposes a joint optimization framework for UAV-mounted Beyond Diagonal Intelligent Reflective Surfaces (BD-IRS) in mobile edge computing (MEC) networks to minimize worst-case system latency. By jointly optimizing UAV deployment, task segmentation, computational resource allocation, power control, and beamforming, the framework reduces latency by 7.25% compared to benchmarks and improves spectral efficiency by 17.77% over conventional diagonal IRS systems.

ABSTRACT

Recent advancements in 6G systems signal a leap towards universal connectivity and ultra-reliable, low-latency communications for real-time data devices. Yet, these advancements encounter obstacles such as limited device battery life and computational power, along with urban signal blockages. To counter these, Intelligent Reconfigurable Surfaces (IRS) within Mobile Edge Cloud (MEC) infrastructures offer enhanced computing to overcome device limitations and create alternative communication paths. Despite these improvements, connectivity issues remain for remote areas. Our paper presents the Beyond Diagonal IRS (BD-IRS or IRS 2.0), integrated with UAVs in MEC networks (BD-IRS-UAV), providing on-demand links for remote users to offload tasks, tackling resource and battery limitations. We propose a joint optimization strategy to reduce system's worst-case latency and UAV hovering time by optimizing BD-IRS-UAV deployment and resource allocation. This challenge is approached by dividing it into two sub-problems: BD-IRS-UAV Placement and Computational Resource Optimization, and Communication Resource Optimization, each solved iteratively. This design significantly enhances system performance, showing a $17.75\%$ increase over traditional diagonal IRS and a $25.43\%$ improvement over IRS on buildings, with a $13.44\%$ enhancement in worst-case latency compared to binary offloading schemes.

Motivation & Objective

• Address the challenge of high latency and limited computational capacity in 6G-enabled mobile edge computing (MEC) networks with massive IoT and data-intensive applications.
• Overcome the limitations of conventional diagonal phase-shift IRS architectures, which restrict passive beamforming flexibility and system performance.
• Introduce a novel BD-IRS-UAV architecture to enhance coverage, capacity, and energy efficiency in MEC networks through dynamic reconfigurability and UAV mobility.
• Minimize worst-case system latency by jointly optimizing UAV positioning, task offloading, computational resource allocation, power control, and beamforming design.
• Evaluate the performance of advanced IRS architectures—fully-connected and group-connected—against conventional single-connected diagonal IRS in MEC scenarios.

Proposed method

• Decompose the non-convex, non-linear, NP-hard optimization problem into two subproblems: (1) UAV deployment, task segmentation, and computational resource allocation; (2) power allocation, beamforming, and phase shift design.
• Transform each subproblem into a convex optimization form using standard techniques such as successive convex approximation (SCA) and D.C. programming.
• Employ alternating optimization to iteratively solve the decoupled subproblems, ensuring convergence to a suboptimal but practical solution.
• Utilize the BD-IRS 2.0 architecture with fully-connected and group-connected topologies to enable non-diagonal, flexible phase shift control for enhanced signal manipulation.
• Integrate task segmentation with partial offloading, where each user’s task is split between local and edge computation based on optimal βn to balance computation and offloading delays.
• Derive a closed-form expression for optimal task segmentation (βn) using the condition that local and offloading computation times are equal, minimizing total delay.

Figure 1 : System model

Experimental results

Research questions

• RQ1 How does the integration of BD-IRS-UAV improve system latency and spectral efficiency compared to conventional diagonal IRS in MEC networks?
• RQ2 What is the impact of different IRS architectures—fully-connected, group-connected, and single-connected—on system performance in terms of latency and data offloading rate?
• RQ3 To what extent does joint optimization of UAV positioning, computational resource allocation, and beamforming reduce worst-case system latency?
• RQ4 How does partial offloading with optimal task segmentation outperform binary offloading or full edge offloading in high-user-density MEC scenarios?
• RQ5 What is the trade-off between performance gain and implementation complexity in fully-connected versus group-connected BD-IRS architectures?

Key findings

• The proposed optimization framework reduces worst-case system latency by 7.25% compared to benchmark frameworks, including binary offloading, full edge offloading, and fixed computation allocation.
• BD-IRS with fully-connected and group-connected architectures achieves a 17.77% higher data rate compared to conventional single-connected diagonal IRS, demonstrating superior spectral efficiency.
• The group-connected BD-IRS architecture achieves performance comparable to the fully-connected design but with significantly lower complexity due to reduced number of phase shift elements.
• As the number of users increases, the performance gain of the proposed partial offloading scheme over binary offloading becomes more pronounced, with latency reduction scaling effectively under congestion.
• System latency is minimized when local and offloading computation times are balanced, which is achieved through the derived optimal task segmentation factor βn.
• Monte Carlo simulations confirm that advanced IRS architectures enable faster data offloading, leading to more efficient computational resource allocation and improved fairness across users with varying task sizes.

Figure 2 : Road-map of iterative algorithms.