"BeamFormer: Ray-Tracing Channel Dataset"

"This dataset contains raw ray-tracing simulation outputs for mmWave beam management research, generated using Wireless InSite for indoor home and office environments at 28 GHz. It is the companion dataset for the paper \"BeamFormer: Transformer-based Beam Management for 6G Networks\" (MobiSys 2026).The dataset comprises 9,853 SQLite scene folders (~300 GB total). Each scene places one fixed transmitter and approximately 100 randomly positioned receivers within an indoor environment, yielding ~100 Tx-Rx channel samples per scene and approximately 900,000+ channel samples in total. Transceiver positions, orientations, and heights are randomized across scenes to maximize spatial diversity.Simulation parameters: center frequency 27.925 GHz, subcarrier spacing 240 kHz, 128 subcarriers (spanning 27.894\u201327.956 GHz). Ray-tracing conditions include up to 6 reflections, 1 diffraction, and 1 transmission per path, with diffuse scattering enabled. Near-field effects are modeled by computing absolute per-element propagation distances.Raw data is provided as SQLite databases (paired tx-v.sqlite and tx-h.sqlite per scene), storing per-path E-field components for both vertical and horizontal polarizations, angle of departure (AoD), angle of arrival (AoA), and interaction geometry. The antenna array size is configurable at processing time: the accompanying open-source MATLAB code (available at [GitHub URL]) reconstructs the full MIMO channel response for any user-specified array geometry and computes beam RSS for arbitrary beamforming vectors. The default configuration used in the paper is a 16\u00d716 URA with 0.5\u03bb spacing."