2nd Workshop on 5G Meets AI/ML: Data-Driven Connectivity, Computing and Control

Session WS5-1

Session 1

Conference
9:00 AM — 10:30 AM CST
Local
Aug 8 Sat, 9:00 PM — 10:30 PM EDT

Deep Reinforcement Learning for Energy Saving in Radio Access Network

Keran Zhang (Beijing University of Posts and Telecommunications, China); Xiang Ming Wen (Beijing University of posts and telecommunications, China); Yawen Chen (Beijing University of Posts and Telecommunications, China); Zhaoming Lu (BUPT, China)

1
Energy consumption of radio access networks (RANs) has become a significant issue, due to the growing density of base stations (BSs). Recent works have validated the possibility of saving energy in RANs by dynamically turning on/off some BSs. However, most of existing works are based on the stochastic traffic models, whose performance is unknown with the real-world traffic. In this paper, we propose a data-driven and deep reinforcement learning based BS sleeping scheme. Specifically, we first build TrafficNet, which is a anonymous traffic data set collected from a commercial RAN in Shanghai. Moreover, we propose a Double Deep Q-Learning Network based BS Sleeping Algorithm (DDQS) to optimize energy utilization while compromising the quality of services (QoS). In addition, to expedite the training process under the TrafficNet, we pre-process the data using manipulation tool to facilitate the data analysis. Simulation results show that the proposed DDQS shows superior performance to the state-of-art sleeping algorithms in terms of energy saving and QoS guarantee.

Vehicle Speed Prediction with Convolutional Neural Networks for ITS

Yifei Li, Celimuge Wu and Tsutomu Yoshinaga (The University of Electro-Communications, Japan)

1
Accurate vehicle speed prediction is one of the important issues for achieving advanced intelligent transportation systems (ITS). There are two main methods for vehicle speed prediction, specifically, the statistical prediction and neural network-based approach. However, most existing neural network-based approaches use recurrent neural networks, which can not address the spatial characteristics of traffic flows. In this paper, we propose a convolutional neural network-based approach for a better estimation of vehicle traffics.

Intelligent Universal Acceleration Framework and Verification for Edge Cloud Applications

Jie Mei (Beijing University of Posts and Telecommunications, China); Bo Lei, Xuliang Wang and Qianying Zhao (Beijing Research Institute China Telcom Beijing, China)

1
With the development of 5G, the applications impose more diverse and higher performance demands on the infrastructure capability of Edge Cloud. In order to meet the demands, many hardware-based Edge Cloud Application Acceleration schemes have been proposed. However, to ensure the vigorous development of Edge Cloud applications, how to efficiently match and schedule the acceleration capability of edge cloud applications with the diverse acceleration demands of edge cloud applications is one of the indispensable prerequisites. In this paper, an intelligent general acceleration framework for Edge Cloud applications is proposed, which mainly solves the following problems: first, distinguish and model the acceleration demands of applications accurately; second, abstractly model and classify the general acceleration capabilities; third, detect the initial period of application acceleration, and provide intelligent policy scheduling framework according to test results. And the framework achieves accurate matching and scheduling between application acceleration demands and general acceleration capabilities. So, the Intelligent Universal Acceleration Framework can accelerate the application of Edge Cloud accurately, and ensure the efficient utilization of resources. In addition, bandwidth-intensive, face recognition and enterprise VPN gateway are used to validates that the Intelligent Universal Acceleration Framework is helpful to analyze, match and schedule the acceleration demands of applications and the general acceleration capabilities of infrastructure.

A Location-aware Computation Offloading Policy for MEC-assisted Wireless Mesh Network

Wenxiao Shi, Sicheng Liu, Jiadong Zhang and Ruidong Zhang (Jilin University, China)

1
Mobile edge computing (MEC), an emerging technology, has the characteristics of low latency, mobile energy savings, and context-awareness. As a type of access network, wireless mesh network (WMN) has gained wide attention due to its flexible network architecture, low deployment cost, and self-organization. The combination of MEC and WMN can solve the shortcomings of traditional wireless communication such as storage capacity, privacy, and security. In this paper, we propose a location-aware (LA) algorithm to cognize the location and a location-aware offloading policy (LAOP) algorithm considering the energy consumption and time delay. Simulation results show that the proposed LAOP algorithm can obtain a higher completion rate and lower average processing delay compared with the other two methods.

Session Chair

Rui Yin

Session WS5-2

Session 2

Conference
11:00 AM — 12:30 PM CST
Local
Aug 8 Sat, 11:00 PM — 12:30 AM EDT

Computation Offloading Scheme with D2D for MEC-enabled Cellular Networks

Minglei Tong and Xiaoxiang Wang (Beijing University of Posts and Telecommunications, China); Yulong Wang (Beijing university of post and telecommunications, China); Yanwen Lan (Beijing University of Posts and Telecommunications, China)

1
With the emergence of computation-intensive mobile applications in wireless networks, the limited computation capacities of some mobile devices (MDs) are not able to satisfy the service requirements and the backhaul of cellular base station (BS) is overloaded. To solve these problems, the combination of mobile edge computation (MEC) technology and device-to-device (D2D) technology is of great concern, which makes MDs have more chances to offload tasks to the network edge, instead of sending them to the remote Clouds, as well as the load of backhaul is reduced effectively. There are several recent studies on combining MEC and D2D technology, but maximizing the number of MDs whose tasks are completed when the tasks are indivisible, and on this basis, reducing the energy consumption of task offloading as far as possible is rarely considered. Aiming to achieve this goal, a computation offloading scheme with D2D for MEC-enabled cellular networks is proposed in this paper, which focus on appropriate task offloading strategies. At first, the optimal problem is transformed to a 0-1 integer programming problem, which can be solved to obtain the maximum number. Then, the energy consumption of task offloading is reduced by exchanging the task offloading strategies between MDs. Finally, numerical simulation results validate that the proposed scheme has relatively better performances than the baseline schemes.

Operation and Security Considerations of Federated Learning Platform Based on Compute First Network

Lei Zhu (China Mobile (Chengdu) Industrial Research Institute, China); Hui Xue (CTO Office & Asiainfo Security, China); Yong Pang (Strategic Advisory Center & Asiainfo Securit, China); Lijun Zhao, Zhengpeng You and Xiaoyong Tang (China Mobile(Chengdu) Industrial Research Institute, China)

1
To provide public federated learning services based on the Compute First Network, the Federated Learning Platform is proposed for communication service providers in this paper. Operation and security are taken into consideration to assure and secure service delivery. The generic framework of Federated Learning Platform is designed to meet the fundamental requirements. Based on the TM Forum Business Process Framework, the specific operation functionalities are introduced. Also, those security issues are addressed by adding a security domain, and a cryptography infrastructure is particularly discussed herein to make connections among the federated communication parties trusted, ensuring mutual authentication and interaction data protection.

Robust Adaptive Beamforming Based on Norm Constraint Regularization Correntropy for Impulsive Interference

Yingke Lei (Electronic Engineering Institute, China)

1
Isotropic beamforming method with dual norm constraints is proposed to reduce the impact of pulsed stable interference and maintain the sparsity of the beamformer weights. The realization of motivation is mainly based on the combination of norm regularization constraints and known maximum correlation criterion (MCC) criterion. First, consider the constraints existing in the constrained least mean square (CLMS) algorithm, combined with MCC to construct a linear constrained optimization equation. In addition, in order to reduce the number of active elements in the limited power array system, improve the directivity performance of the beamformer to improve the steadystate sparsity. We introduce a norm penalty equation in the constraint list of the adaptive filter, so that the smaller coefficient is zero and highlights the globally optimal optimization area. The convexity and stability of the norm can eliminate the instability of sparse performance. The convergence and complexity of the algorithm are analyzed, and the stability conditions of convergence are given. Simulation results show that the proposed method is superior to other existing beamforming techniques.

Session Chair

Rui Yin

Made with in Toronto · Privacy Policy · © 2020 Duetone Corp.