Lukas Eller

Machine learning for mobile network analysis, planning, and optimization is often limited by the lack of large, comprehensive real-world datasets. This paper introduces the Vienna 4G/5G Drive-Test Dataset, a city-scale open dataset of georeferenced Long Term Evolution (LTE) and 5G New Radio (NR) measurements collected across Vienna, Austria. The dataset combines passive wideband scanner observations with active handset logs, providing complementary network-side and user-side views of deployed radio access networks. The measurements cover diverse urban and suburban settings and are aligned with time and location information to support consistent evaluation. For a representative subset of base stations (BSs), we provide inferred deployment descriptors, including estimated BS locations, sector azimuths, and antenna heights. The release further includes high-resolution building and terrain models, enabling geometry-conditioned learning and calibration of deterministic approaches such as ray tracing. To facilitate practical reuse, the data are organized into scanner, handset, estimated cell information, and city-model components, and the accompanying documentation describes the available fields and intended joins between them. The dataset enables reproducible benchmarking across environment-aware learning, propagation modeling, coverage analysis, and ray-tracing calibration workflows.

The efficient operation of cellular networks requires precise tuning of configuration parameters such as the antenna downtilt or the transmit power. Data-driven methods, which can integrate feedback from monitoring data, are promising but face challenges related to sample efficiency, scalability, and safety, limiting their real-world application. In this work, we introduce an innovative online coverage and capacity optimization framework that combines model-free exploration using Monte Carlo tree search with a safe, model-based baseline derived from a probabilistic differentiable network twin. We formulate the optimization task as a sequential tree search problem, developing specialized policies to guide the exploration of the configuration space. The differentiable network twin aids both the selection policy, by pruning the search space with a prior action distribution, and the rollout policy, enabling domain knowledge-based selection of remaining actions. Our results demonstrate that the proposed approach effectively guides the optimization process, outperforming both purely model-free and model-based methods. Our solution improves safety by reducing the risk of testing poorly performing configurations, enhances the model-based solution, and can also compensate for severe model mismatches in the digital twin. The framework thus addresses a common obstacle in applying data-driven optimization to real-world network deployments.

The efficient operation of cellular networks requires precise tuning of configuration parameters such as the antenna downtilt or the transmit power. Here, data-driven methods, which can integrate feedback from monitoring data, are promising but face challenges related to sample efficiency, scalability, and safety --- limiting their real-world application. In this work, we introduce an innovative online coverage and capacity optimization framework that combines model-free exploration using Monte Carlo tree search with a safe, model-based baseline derived from a probabilistic differentiable network twin. We formulate the optimization task as a sequential tree search problem, developing specialized policies to guide the exploration of the configuration space. The differentiable network twin aids both the selection policy, by pruning the search space with a prior action distribution, and the rollout policy, enabling domain knowledge-based selection of remaining actions. Our results demonstrate that the proposed approach effectively guides the optimization process, outperforming both purely model-free and model-based methods. Our solution improves safety by reducing the risk of testing poorly performing configurations, enhances the model-based solution, and can also compensate for severe model mismatches in the digital twin. Hence, the framework addresses a common obstacle in applying data-driven optimization to real-world network deployments.

The efficient operation of cellular networks requires careful tuning of configuration parameters, such as the transmit power or antenna tilts, to adequately balance interference while providing the necessary capacity to the connected UEs. As manual tuning of these parameters is typically unfeasible, several automated coverage and capacity optimization methods have been proposed. However, most existing solutions are either based on poorly scalable black-box optimization methods or solely consider interference management, while omitting the potential of congested cells. In this work, we instead propose a differentiable framework for cellular network optimization, centered around the end-user throughput, that enables load-aware tuning of network parameters through gradient descent. Hereby, we approach the problem from a data-driven perspective, and include dedicated model subcomponents derived from monitoring data, which enable the calibration to site-specific traffic patterns and KPI measurements. We validate our approach for joint transmit power optimization in a real-world network layout with $\approx 150$ cells in two frequency bands. In our evaluation, the gradient descent-based optimization reliably reduces the outage ratio for different levels of demand, while the black-box baseline struggles to explore the large search space. Our results further reveal substantial differences between the proposed load-aware and commonly used SINR-based objectives, for which we repeatably obtain unbalanced network configurations with severely congested cells. In contrast, the proposed end-user throughput objective promotes a balanced network configuration, providing adequate resources to the connected UEs while also limiting interference.

Cellular operators tightly monitor their networks to keep up with the market demand and frequently benchmark their performance against competitors. Typical benchmarking tests compare Key Performance Indicators (KPIs), Quality of Service (QoS), and Quality of Experience (QoE) parameters on the city- and regional levels using user-collected crowdsourced data or drive test measurements. However, time-variant parameters and different user mobility patterns can bias the performance comparison. Designing a measurement sampling strategy that deals with such issues is critical for achieving a valid benchmark. Whether we would like to determine how many tiles of a map have to be measured in drive tests or how many samples we need from crowdsourced data to reach an estimate with the required accuracy, sampling theory can provide us with an answer. Since propagation conditions depend on user mobility and measurement environment, splitting the data set into groups or strata allows us to attain an unbiased estimate with fewer samples, thus allowing for a fair comparison to other Mobile Network Operators (MNOs) with minimum effort measurements. In this work, we characterize the performance of different sampling methods on the simulated KPI data set while investigating specific use cases to reveal scenarios where the stratification method pays off. We analyze the sampling methods on two real Minimization of Drive Test (MDT) crowdsourced data sets from a major Austrian operator. By stratifying the data into meaningful strata, we obtain the required number of areas and measurements in each stratum while remaining under the pre-set estimation error level. To our knowledge, this is the first study on sampling methodologies applied to real-world crowdsourced cellular measurements.

In urban scenarios, network planning requires awareness of the notoriously complex propagation environment by accounting for blocking, diffraction, and reflection on buildings. To this end, deep learning-based signal-strength prediction directly operating on environmental data has recently gained attention, mainly as a computationally efficient alternative to ray-tracing. Our work combines RSRP measurements from an extensive drive-test campaign in a live 4G network with a 3D city model for the largest real-world assessment of such data-driven schemes to date. We compare three different encodings of the propagation environment and find that a neural network operating on a full 3D representation of the surroundings performs best with an RMSE of 7.06 dB . It is followed by a model using only the direct path profile with 7.78 dB and a reference neural network utilizing a binary line of sight indicator achieving 8.76 dB . The large size of our data set allows us to address several open questions regarding the inner workings of these black box approaches. In particular, we elaborate on different evaluation strategies, highlighting the importance of spatial separation of train and test areas, as the rich environmental data implicitly provides a spatial reference. Through model explainability, we further identify the area along the direct path between the user equipment and the transmitter as the input region with the highest feature importance — questioning the common practice of including large buffer areas. Evaluating the models in scenarios with artificially placed base stations reveals that the measurement campaign offers a sufficient basis for a prototypical network planner. The trained models, which we make publicly available, exhibit the dominant propagation mechanisms in urban areas and generate spatially consistent and physically sound signal-strength maps.

Although mobile communication has become ubiquitous in our modern society, operators typically treat the underlying networking infrastructure in a secretive manner. However, detailed topology information is a key enabler for operator benchmarking and can serve as ground-truth data for system-level simulations or user equipment localization schemes. Still, existing approaches for base station localization that are either based on received signal strength or on the spatial distribution of measurements are not accurate enough for such use cases. Accordingly, we propose a localization scheme that operates on end-user measurements of the 4G & 5G timing advance parameter, which acts as a quantized distance measure between the user equipment and the base station. By directly incorporating GPS noise, multipath propagation, and quantization into our stochastic system model, we obtain an estimator that offers a reliable measure of confidence and requires only the configuration of two hyperparameters with a dedicated physical interpretation. We evaluate our approach using a set of drive-test measurements consisting of 190 LTE eNodeBs with ground-truth locations confirmed by an Austrian mobile network operator. Our selective estimator can either operate without prior knowledge, resulting in mean distance errors of below 100 m, or in a classification setup, where it correctly identifies up to 95% of eNodeBs from a set of candidate cell tower locations. To allow for reproducibility, we make our dataset and a reference implementation publicly available.