The inherent flexibility of 5G networks come with a high degree of configuration and management complexity. This makes the performance outcome for UEs, more than ever, dependent on intricate configurations and interplay between algorithms at various network components. In this paper, we take initial steps towards a performance exposure system at the base station using a data-driven approach for predicting performance violations in terms of RTT, as observed by the UE, in a 5G mmWave network. We present ML models to predict RTT using low-level and high-frequency base station metrics from a 5G mmWave testbed based on commercially available equipment. Predicting UE performance from a base station perspective, and exposing this knowledge, is valuable for applications to proactively address performance violations. We also compare several methods for feature reduction, which have a significant impact on monitoring load. We demonstrate our model's ability to identify RTT violations, paving the way for network providers towards an intelligent performance exposure system. 

Dynamic Adaptive Streaming over HTTP (DASH) is a standard for delivering video in segments and adapting each segment's bitrate (quality), to adjust to changing and limited network bandwidth. We study segment prefetching, informed by machine learning predictions of bitrates of client segment requests, implemented at the network edge. We formulate this client segment request prediction problem as a supervised learning problem of predicting the bitrate of a client's next segment request, in order to prefetch it at the mobile edge, with the objective of jointly improving the video streaming experience for the users and network bandwidth utilization for the service provider. The results of extensive evaluations showed a segment request prediction accuracy of close to 90% and reduced video segment access delay with a cache hit ratio of 58%, and reduced transport network load by lowering the backhaul link utilization by 60.91%.

From a 5G operator's perspective, accurate estimates of key User Equipments (UEs) performance metrics, especially One-Way Delay (OWD), can provide valuable information. These estimates can trigger management tasks such as reconfiguration to prevent violations of Service Level Objectives (SLOs). Moreover, such insights into UE performance can empower applications to adapt their services to end-users in a more effective manner. We use advanced machine learning over data gathered at the base stations to predict OWD from UEs and show that we are able to predict OWD with over a 2x reduction in percentage error compared to the considered baseline. We discover the close coupling between the performance of the OWD model and the type of UE, which poses a model generalization challenge. Addressing this problem, we demonstrate the shortcomings of the commonly used fine-tuning approach and develop a novel method based on domain adversarial neural networks, that can adapt to a target domain without compromising on the performance of the source domain. Our results show that we can adapt our source model to provide OWD prediction performance within 1-4 percentage points of the ideal scenario when the source and the target domains are the same. Also, our work is grounded in empirical experiments conducted within a 5G testbed, using commercially available hardware.

Predicting user performance at the base station in telecom networks is a critical task that can significantly benefit from advanced machine learning techniques. However, labeled data for user performance are scarce and costly to collect, while unlabeled data consisting of base station metrics, are more readily accessible. Self-supervised learning provides a means to leverage this unlabeled data, and has seen remarkable success in the domains of computer vision and natural language processing, with unstructured data. Recently, these methods have been adapted to structured data as well, making them particularly relevant to the telecom domain. We apply self-supervised learning to predict user performance in telecom networks. Our results demonstrate that even with simple self-supervised approaches, the percentage of variance of the target values explained by the model, in low-labeled scenarios (e.g., only 100 labeled samples) can be improved fourfold, from 15% to 60%. Moreover, to promote reproducibility and further research in the domain, we open-source a dataset creation framework and a specific dataset created from it that captures scenarios that have been deemed to be challenging for future networks.