This paper presents a real-world dataset intended to help machine learning for mobile networks. The authors collected radio and signaling data from a commercially deployed network while users moved at different speeds. Their main goal is to enable AI methods that reduce handover interruption time and keep throughput steady when a device switches from one cell to another.

The team recorded data for several modes of movement: pedestrian, bike, car, bus, and train. They focused on handover events — the moments when a user device changes its serving cell — and on beam management, which is how the network chooses directional radio beams. Importantly, the dataset includes timing advance (TA) measurements. Timing advance is a short number that tells the network how far a device is, so signals line up in time. The TA values were logged at specific signaling events such as RACH triggers (Random Access Channel), MAC CE (Medium Access Control control element), and PDCCH grants (Physical Downlink Control Channel). These event-level TA measurements are often missing from other datasets.

Besides collecting the raw data, the paper describes how the dataset was created. The authors provide details on the experimental setup, how they captured and extracted the records, and an initial exploratory analysis. That analysis looks at mobility behavior, beam selection patterns, and timing advance trends during handovers. The paper also outlines use cases for the dataset, for example training and testing AI models that predict timing advance or help decide when and how to switch beams or cells.

Why this matters: much prior work in AI for mobility and beam management relies on simulations. Simulated data can miss real deployment quirks and real user traffic patterns. A dataset gathered from a live commercial network can give machine learning models more realistic examples. Better models could lead to shorter service interruptions when you move and to steadier data rates during and after handovers.