OPNet-Sim: A synthetic benchmark dataset for multi-city 5G network performance and user experience modelling

Abstract

The global rollout of 5G technology promises unprecedented data rates, ultra-low latency, and massive device connectivity. However, the research community often lacks access to large-scale, real-world datasets needed to model the highly heterogeneous nature of network performance and user experience (QoE). A complex interplay of radio frequency conditions, network deployment strategies, and device capabilities shapes these characteristics. While traditional drive-testing can provide granular data, its utility is limited by spatial and temporal constraints, making it unsuitable for continuous large-scale analysis. To address this data gap, this paper introduces OPNet-Sim, a framework for generating realistic, large-scale, multi-dimensional synthetic datasets that emulate data collected from commercial 5G smartphones. The design of OPNet-Sim is informed by statistical characteristics and data schemas found in the literature and public reports on large-scale network measurement. The simulated dataset encompasses over 1.2 billion synthetic records, emulating data from more than 150,000 unique devices over 12 months. It includes detailed physical layer measurements (e.g., RSRP, RSRQ, SINR), key performance indicators (KPIs) such as throughput and latency, device context information, and network metadata. OPNet-Sim serves as both a benchmark and a synthetic data resource for researchers in telecommunications and data science. It enables the development, training, and validation of models for network performance prediction, QoE estimation for applications such as video streaming, and novel methodologies for network diagnostics all without the privacy and access constraints associated with real user data. This paper describes the dataset generation methodology, the structural schema, validation against established models, and illustrative examples of potential applications.