The emergence of novel cellular network technologies, within 5G, are envisioned as key enablers of a new set of use-cases, including industrial automation, intelligent transportation, and tactile internet. The critical nature of the traffic requirements ranges from ultra-reliable communications, massive connectivity, and enhanced mobile broadband. Thus, the growing research on cellular network monitoring and prediction aims for ensuring a satisfied user-base and fulfillment of service level agreements. The scope of this study is to develop an approach for predicting the cellular link throughput of end-users, with a goal to benchmark the performance of network slices. First, we report and analyze a measurement study involving real-life cases, such as driving in urban, sub-urban, and rural areas, as well as tests in large crowded areas. Second, we develop machine learning models using lower-layer metrics, describing the radio environment, to predict the available throughput. The models are initially validated on the LTE network and then applied to a non-standalone 5G network. Finally, we suggest scaling the proposed model into the future standalone 5G network. We have achieved 93% and 84% R^2 accuracy, with 0.06 and 0.17 mean squared error, in predicting the end-user's throughput in LTE and non-standalone 5G network, respectively.

Connected and automated vehicles (CAVs) are envisioned to revolutionize the transportation industry, enabling autonomous processes and real-time exchange of information among vehicles and infrastructure. To safely navigate the roadways, CAVs rely on sensor readings and data from the surrounding vehicles. Hence, a fault or anomaly arising from the hardware, software, or the network can lead into devastating consequences regarding safety. This study investigates potential performance degradation caused by anomalies, by analyzing real-life vehicles’ sensory and network-related data. The aim is to utilize unsupervised learning for anomaly detection, with a goal to describe the cause and effect of the detected anomalies from a performance perspective. The results show around 93% F1-score when detecting anomalies imposed by the cellular network and the vehicle’s sensors. Moreover, with approximately 90% F1-score we can detect anomalous predictions from a deployed network-related ML model predicting cellular throughput and describe the root-causes behind the detected anomalies.

Today's research on Quality of Experience (QoE) mainly addresses multimedia services. With the introduction of the Internet of Things (IoT), there is a need for new ways of evaluating the QoE. Emerging IoT services, such as autonomous vehicles (AVs), are more complex and involve additional quality requirements, such as those related to machine-to-machine communication that enables self-driving. In fully autonomous cases, it is the intelligent machines operating the vehicles. Thus, it is not clear how intelligent machines will impact end-user QoE, but also how end users can alter and affect a self-driving vehicle. This article argues for a paradigm shift in the QoE area to cover the relationship between humans and intelligent machines. We introduce the term Quality of IoT-experience (QoIoT) within the context of AV, where the quality evaluation, besides end users, considers quantifying the perspectives of intelligent machines with objective metrics. Hence, we propose a novel architecture that considers Quality of Data (QoD), Quality of Network (QoN), and Quality of Context (QoC) to determine the overall QoIoT in the context of AVs. Finally, we present a case study to illustrate the use of QoIoT.

The use of video streaming services are increasing in the cellular networks, inferring a need to monitor video quality to meet users’ Quality of Experience (QoE). The so-called no-reference (NR) models for estimating video quality metrics mainly rely on packet-header and bitstream information. However, there are situations where the availability of such information is limited due to tighten security and encryption, which necessitates exploration of alternative parameters for conducting video QoE assessment. In this study we collect real-live in-smartphone measurements describing the radio link of the LTE connection while streaming reference videos in uplink. The radio measurements include metrics such as RSSI, RSRP, RSRQ, and CINR. We then use these radio metrics to train a Random Forrest machine learning model against calculated video quality metrics from the reference videos. The aim is to estimate the Mean Opinion Score (MOS), PSNR, Frame delay, Frame skips, and Blurriness. Our result show 94% classification accuracy, and 85% model accuracy (R2value) when predicting the MOS using regression. Correspondingly, we achieve 89%, 84%, 85%, and 82% classification accuracy when predicting PSNR, Frame delay, Frame Skips, and Blurriness respectively. Further, we achieve 81%, 77%, 79%, and 75% model accuracy (R2value) regarding the same parameters using regression.

Industrial Internet of Things (IIoT) is claimed to be a global booster technology for economic development. IIoT brings bulky use-cases with a simple goal of enabling automation, autonomation or just plain digitalization of industrial processes. The abundance of interconnected IoT and CPS generate additional burden on the telecommunication networks, imposing number of challenges to satisfy the key performance requirements. In particular, the QoS metrics related to real-time data exchange for critical machine-to-machine type communication. This paper analyzes a real-world example of IIoT from a QoS perspective, such as remotely operated underground mining vehicle. As part of the performance evaluation, a software tool is developed for estimating the absolute, one-way delay in end-to-end transmissions. The measured metric is passed to a machine learning model for one-way delay prediction based on LTE RAN measurements using a commercially available cutting-edge software tool. The achieved results prove the possibility to predict the delay figures using machine learning model with a coefficient of determination up to 90%.