Multi-Operator Dataset for Throughput Prediction Across Diverse Mobility Modes

Description

This dataset features real-world radio frequency (RF) measurements and network performance metrics collected within a dense urban and academic ecosystem in Sunway City, Malaysia. The data was engineered to support research in mobile network performance, user equipment (UE) behavior, and the predictive modeling of user-level throughput in 5G New Radio (NR) and LTE environments. The data collection architecture utilized commercial Samsung Galaxy S24 smartphones interfaced with the Keysight Nemo Handy drive test tool to capture high-fidelity engineering logs directly from the baseband processor. The dataset uniquely captures concurrent performance metrics across three top-tier mobile network operators. To facilitate research into realistic user experiences, data collection spanned three distinct micro-mobility environments reflective of modern smart-city infrastructure: Canopy Walks: Representing low-speed pedestrian mobility and elevated walkways. Shuttle Buses: Representing medium-speed street-level transit. Elevated BRT (Bus Rapid Transit): Representing dedicated, higher-frequency elevated transit corridors. Records include highly granular, sub-second tracking of spatial parameters (Latitude, Longitude), radio link quality (RSRP, Band, Physical Cell Identity), and network layer attributes (Beam Index, Cell Type). The primary target variable for machine learning frameworks and time-series forecasting is Throughput_Mbps, captured during active network sessions.

Steps to reproduce

Field measurements were conducted by researchers navigating pre-defined transit routes (canopy walks, street-level shuttle bus lines, and an elevated BRT line). Network diagnostic logging was performed using Keysight Nemo Handy software installed on commercial Samsung Galaxy S24 handsets, maintaining continuous active data transfer sessions to log real-time application-layer download throughput. The resulting raw .nmf (Nemo File Format) log files were post-processed and parsed using Nemo analyze software. Key physical layer RF parameters, beam indices, and geographic coordinates were extracted, synchronized chronologically at intervals, and structured into this unified tabular CSV format for machine learning application.

Institutions

• Sunway University

  Selangor, Petaling Jaya

Categories

Licence