[Paper Review] Reconfigurable Intelligent Surface Assisted Mobile Edge Computing with Heterogeneous Learning Tasks

Shanfeng Huang, Shuai Wang+3 more|arXiv (Cornell University)|Dec 25, 2020

Advanced Wireless Communication Technologies50 references4 citations

TL;DR

This paper proposes an RIS-assisted mobile edge computing framework to optimize heterogeneous machine learning tasks by jointly minimizing the maximum learning error across users through power control, beamforming, and RIS phase shifts. It introduces an AO-based algorithm with SCA, ADMM, and error-level search to solve the nonconvex optimization, achieving significant performance gains over benchmarks in 3D object detection use cases.

ABSTRACT

The ever-growing popularity and rapid improving of artificial intelligence (AI) have raised rethinking on the evolution of wireless networks. Mobile edge computing (MEC) provides a natural platform for AI applications since it is with rich computation resources to train machine learning (ML) models, as well as low-latency access to the data generated by mobile and internet of things (IoT) devices. In this paper, we present an infrastructure to perform ML tasks at an MEC server with the assistance of a reconfigurable intelligent surface (RIS). In contrast to conventional communication systems where the principal criterions are to maximize the throughput, we aim at maximizing the learning performance. Specifically, we minimize the maximum learning error of all participating users by jointly optimizing transmit power of mobile users, beamforming vectors of the base station (BS), and the phase-shift matrix of the RIS. An alternating optimization (AO)-based framework is proposed to optimize the three terms iteratively, where a successive convex approximation (SCA)-based algorithm is developed to solve the power allocation problem, closed-form expressions of the beamforming vectors are derived, and an alternating direction method of multipliers (ADMM)-based algorithm is designed together with an error level searching (ELS) framework to effectively solve the challenging nonconvex optimization problem of the phase-shift matrix. Simulation results demonstrate significant gains of deploying an RIS and validate the advantages of our proposed algorithms over various benchmarks. Lastly, a unified communication-training-inference platform is developed based on the CARLA platform and the SECOND network, and a use case (3D object detection in autonomous driving) for the proposed scheme is demonstrated on the developed platform.

Motivation & Objective

• Address the challenge of optimizing learning performance in RIS-assisted mobile edge computing, where traditional throughput-centric designs are suboptimal.
• Formulate a joint optimization problem to minimize the maximum learning error across heterogeneous users with varying model and data complexities.
• Design an alternating optimization framework to handle the nonconvex, coupled optimization of transmit power, beamforming vectors, and RIS phase shifts.
• Develop specialized algorithms—SCA for power allocation, closed-form beamforming, and ADMM with error-level search for phase shift optimization.
• Demonstrate the practical viability of the framework through a unified communication-training-inference platform and a real-world 3D object detection use case.

Proposed method

• Propose an alternating optimization (AO) framework to iteratively optimize three variables: user transmit power, base station beamforming vectors, and RIS phase-shift matrix.
• Use successive convex approximation (SCA) to transform the nonconvex power allocation problem into a sequence of convex subproblems.
• Derive closed-form expressions for the optimal beamforming vectors by solving a generalized eigenvalue problem based on SINR maximization.
• Design an ADMM-based algorithm combined with an error level searching (ELS) framework to efficiently solve the nonconvex phase-shift matrix optimization.
• Integrate the algorithms into a unified platform built on CARLA and SECOND network for end-to-end validation of the proposed scheme.
• Model the learning performance as a function of signal-to-interference-plus-noise ratio (SINR), linking communication quality directly to learning accuracy.

Experimental results

Research questions

• RQ1 How can RIS be leveraged to improve the learning performance of heterogeneous machine learning tasks in mobile edge computing?
• RQ2 What joint optimization framework can effectively minimize the maximum learning error across users with diverse model and data characteristics?
• RQ3 How can the nonconvex optimization problem involving transmit power, beamforming, and RIS phase shifts be efficiently solved with convergence guarantees?
• RQ4 What is the performance gain of deploying an RIS in edge learning systems compared to conventional MEC without RIS?
• RQ5 Can a practical end-to-end platform be built to demonstrate real-time communication, training, and inference in RIS-assisted edge AI?

Key findings

• The proposed AO-based algorithm with SCA, closed-form beamforming, and ADMM-ELS achieves convergence and significantly improves learning performance.
• Simulation results show that deploying an RIS leads to substantial gains in minimizing the maximum learning error across users compared to benchmark schemes.
• The closed-form beamforming solution is derived as the dominant eigenvector of a matrix involving channel and interference statistics.
• The ADMM-ELS framework effectively handles the nonconvex phase-shift optimization, enabling practical implementation of RIS in edge learning.
• The end-to-end platform demonstrates successful 3D object detection in autonomous driving, validating the real-world applicability of the proposed framework.
• The objective value of the optimization problem is non-increasing across AO iterations, proving convergence of the proposed algorithm.