Forecasting Quality of Service for Next-Generation Data-Driven WiFi6 Campus Networks


Ak E., Canberk B.

IEEE TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT, vol.18, no.4, pp.4744-4755, 2021 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 18 Issue: 4
  • Publication Date: 2021
  • Doi Number: 10.1109/tnsm.2021.3108766
  • Journal Name: IEEE TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC
  • Page Numbers: pp.4744-4755
  • Keywords: Quality of service, Forecasting, Wireless fidelity, Next generation networking, Wireless networks, Predictive models, Correlation, QoS forecasting, WiFi, graph convolution networks, data-driven networking, wireless networks
  • Istanbul Technical University Affiliated: Yes

Abstract

Forecasting the users' movements and behaviors is extremely valuable for early warning systems to provide high-quality service in wireless and cellular networks. However, forecasting the service of the specific network devices with the additional knowledge of user behaviors is underexplored. This study proposes a WiFi6-specific QoS forecasting engine, which uses a spatio-temporal graph approach to predict QoS parameters, e.g., throughput, in terms of user position in WiFi6 networks. Since WiFi6 networks are planning to meet various traffic types with dense users, it is crucial to analyze it in both spatial and temporal manner by preserving graph-structured data. In this study, we modeled the problem with a novel deep learning approach, Graph Convolution Networks (GCNs), by adapting the Omni-Scale 1D CNN for temporal analysis. Then, we analyze the forecasting performance with two datasets in terms of variety error metrics over loss rate, link speed, throughput, and round trip time (RTT). Also, we give the baselines, ARIMA, FARIMA, SVR, and RNN to compare the proposed solution in terms of accuracy. Finally, we present the simulation results to compare the proposed QoS forecasting approach with user mobility forecasting. All experiments show that proposed WiFi6-specific QoS forecasting gives superior results for multi-horizon QoS prediction with respect to user positions considering heterogeneous traffic types.