The ability to forecast mobile traffic patterns is key to resource management for mobile network operators and planning for local authorities. Several Deep Neural Networks (DNN) have been designed to capture the complex spatio-temporal characteristics of mobile traffic patterns at scale. These models are complex black boxes whose decisions are inherently hard to explain. Even worse, they have proven vulnerable to adversarial attacks which undermine their applicability in production networks. In this paper, we conduct a first in-depth study of the vulnerabilities of DNNs for large-scale mobile traffic forecasting. We propose DeExp, a new tool that leverages EXplainable Artificial Intelligence (XAI) to understand which Base Stations (BSs) are more influential for forecasting from a spatio-temporal perspective. This is challenging as existing XAI techniques are usually applied to computer vision or natural language processing and need to be adapted to the mobile network context. Upon identifying the more influential BSs, we run state-of-the art Adversarial Machine Learning (AML) techniques on those BSs and measure the accuracy degradation of the predictors. Extensive evaluations with real-world mobile traffic traces pinpoint that attacking BSs relevant to the predictor significantly degrades its accuracy across all the scenarios.

The characterization of user behavior is key to perform traffic analysis, modeling and optimization of network components and protocols. This is especially true for 5G and beyond networks that will heavily rely on machine learning for network optimization. In Base Station (BS) traffic traces, users are uniquely identified by a Radio Network Temporary Identifier (RNTI) assigned to them. RNTIs are not bound to a user but are reused upon expiration of an inactivity timer, whose duration is operator dependent. This implies that, over time, multiple users can be mapped to the same RNTI ID in diverse ways. Hence, when using real-world radio access measurement traces for traffic analysis, distinguishing individual users within the RNTI space is a non-trivial task. In this paper, we provide the first in-depth study of the RNTI allocation process and shed light not only on the setting of the inactivity timer, but also on the relationship of the RNTI allocation scheme and the user characteristics. For this, we collect a large dataset of mobile traffic from multiple BSs of several mobile network operators. The analysis of the decoded control messages of the BS unveils that the RNTI allocation process changes over time depending on the BSs observed load and time of day. We also observe that the RNTI expiration threshold is on the order of minutes, and demonstrate how using thresholds around 10~s that are reported in the vast majority of the literature can bias subsequent analyses. Overall, our work provides an important step towards dependable mobile network trace analysis, and lays more solid foundations to research relying on traffic traces for data-driven analysis and simulation.

Beyond 5G mobile networks have to support a wide range of performance requirements and unprecedented levels of flexibility. To this end, massive MIMO is a critical technology to improve spectral efficiency and thus scale up network capacity, by increasing the number of antenna elements. This also increases the overhead of Channel State Information (CSI) estimation and obtaining accurate CSI is a fundamental problem in massive MIMO systems. In this paper, we focus on scheduling uplink Sounding Reference Signals (SRSs) that carry pilot symbols for CSI estimation. Under the large number of users and high load that are expected to characterize beyond 5G systems, the limited amount of resources available for SRSs makes the legacy 3GPP periodic allocation scheme largely inefficient. We design TRADER, an SRS resource allocation framework that minimizes the age of channel estimates by taking advantage of machine learning-based short-term traffic forecasts at the base station level. By anticipating traffic bursts, TRADER schedules SRS resources so as to obtain CSI for each user right before the corresponding traffic arrives. Experiments with extensive real-world mobile network traces show that our solution is efficient and robust in high load scenarios: with respect to a round robin schedule of aperiodic SRS, TRADER provides more often CSI within the coherence time (up to 5× for given scenarios), leading to channel gains of up to 2 dB.

Event-based vision fosters a new way of sensing reality. Event-based cameras work radically differently compared to legacy frame-based cameras because they continuously measure brightness changes at a per-pixel granularity (i.e., events) rather than snapshots of intensity measurements (i.e., frames). Event-based cameras are applied in robotics and augmented and virtual reality applications due to their properties of low-latency, high temporal resolution and dynamic range. For example, they greatly improve unmanned aerial vehicle (UAV) navigation and collision avoidance. While event-based vision is currently restricted to local devices, in the near future applications involving distributed systems will gain momentum, such as the coordination of swarms of UAVs or robots. However, the network traffic characteristics of event-based vision systems are largely unexplored. In this paper, we aim to fill this gap by providing the first study of network traffic generated by event-based cameras. To this end, we employ publicly available data sets and experimentally study properties like the impact of packet/event losses on typical computer vision operations like tracking, and the implications of medium access under contention. We find that complex scenes that incur a high event generation rate are more robust against packet loss due to transmission errors or wireless contention. Conversely, packet loss or delay are more harmful to tracking and visualization operations when the event generation rate is small.