ASR - Adaptive Similarity-Based Regressor for Uplink Data Rate Estimation in Mobile Networks

Georgi Nikolov,Bernd-Ludwig Wenning,Michael Kuhn,Alan Mcgibney

doi:10.1109/jsac.2020.3000414

Abstract

This paper presents a passive data rate estimation method that leverages commonly available parameters of commercial modems with application in Intelligent Transportation Systems. The estimation is performed by utilizing an Adaptive Similarity-based Regression (ASR) approach. This constitutes the use of Support-Vector Regression (SVR) in conjunction with a similarity-based unlearning algorithm. It is demonstrated that this approach can adapt to the various properties of different mobile networks, while maintaining a fixed training set size. This is particularly useful in cases where training data is not available in large quantities, or the uplink rate is limited by the users subscription. ASR is developed as a set of modular components and as such can be used as an enhancement to protocols such as multipath Transmission Control Protocol (TCP), Software-Defined Networking (SDN), or as an aid to Quality-of-Service (QoS) routing. It is shown that the algorithm can achieve satisfactory performance with as little as 24 training samples, and can be deployed across different mobile networks without the need of pre-training. The solution is validated using a custom test-bed to perform mobile network measurements, gathering over 15, 000 measurement samples. In addition, the algorithm is tested using measurements collected under real life conditions both in a moving car and train. As there is a shortage of open source data in the field of rate estimation in mobile networks, we publish all of the data sets used in this paper to encourage further research on the subject.

Highlights

M ANAGING heterogeneous networks presents a challenging task and is still a significant area of research
Offline algorithms applied in previous work [6] are used as a basis to illustrate how the Adaptive Similarity-based Regression (ASR) approach can outperform traditional offline learning techniques due to its ability to rapidly adapt to the dynamics in the network
The use of large training sets for the Neural Network (NN) is necessary, as it has been empirically shown that their performance improves as more data is added to the training set [30]

Summary

Introduction

M ANAGING heterogeneous networks presents a challenging task and is still a significant area of research. With the proliferation of Internet of Things (IoT) applications, gateway mediated architectures are becoming more common place, i.e. sensors, actuators and services deployed in the physical world utilise a gateway to access cloud services and enterprise applications. Gateways bridge the gap between the physical and digital worlds and are used. Manuscript received October 1, 2019; revised February 15, 2020; accepted March 31, 2020. Date of publication June 5, 2020; date of current version September 16, 2020. Michael Kuhn is with the Department of Electrical Engineering and Information Technology, Darmstadt University of Applied Sciences, 64295 Darmstadt, Germany

Objectives

Results

Conclusion