Cross-layer Dataset for Evaluating the Impact of Antenna Placement on Vehicular Cellular Connectivity using POWDER Platform

Problem description: Antenna placement on a vehicle can substantially affect wireless connectivity because the vehicle body may attenuate signals, and different locations create distinct multipath conditions. These effects become especially important during turns, when the antenna’s geometry relative to the base station changes rapidly. Why cross-layer information helps: Synchronized radio and transport-layer measurements enable cross-layer analysis of how fluctuations in RSRP, RSRQ, and SINR translate into changes in throughput, latency, and packet loss. Testbed setup: We have deployed two adjacent cells on the rooftop of the Merrill Engineering Building (MEB) in the POWDER platform at the University of Utah. We use a Quectel RM UE with roof-mounted 5G cellular antennas and drive around the MEB parking lot. Data Collection and Cross-Layer Measurements: We collect data and compare the performance of two antenna placement scenarios: rear-middle placement and front-middle placement. The placements are shown in the attached antenna placement image. We collect the following metrics for cross-layer analysis between UE and gNB. Downlink RSRP, RSRQ, and SINR: Collected by 2 sources, both measuring the Synchronization Signal Block (SSB): KeySight Nemo: Measures at a coarse granularity (1-2 times a second) gNB (OCUDU) logs: RRC measurement reports sent at a fine granularity (every 120ms) Throughput, latency, packet loss: Collected in iperf3 logs measured at the UE end (receiver) Latitude, Longitude Captured in KeySight Nemo logs. Analysis Results: The analysis figure shows the time-indexed difference in RSRP (dBm) and DL bitrate (Mbps) between rear-middle and front-middle placement. The rear-middle placement provides a 7.53 dBm lower median RSRP difference, while the front-middle placement achieves 10.25 Mbps higher throughput. The reason is that DL SS-RSRP has only a weak correlation with downlink bitrate because it is measured on the Broadcast channel (PBCH), whereas downlink application data is transmitted over the Downlink shared channel (PDSCH). Additional context: The measurement and analysis is a basic "end-to-end" example of data collection and processing using the POWDER Data and Model Commons framework. The framework aims to simplify the collection, storage, and processing of data from multiple data sources in a distributed wireless "experiment". As described above, the data sources here include network data (OCUDU logs), data collected by a UE based tool (KeySight Nemo), and application level data (iperf logs). Data is stored in a data lake and the framework provides a client that streams data from different data sources in the POWDER experiment into the data lake. The framework also makes it easy to associate user defined meta-data with the data collected across all data sources. E.g., in this case with two different user defined "runs" according to the antenna placement. Finally, as described above, the example includes two analysis scripts which extract and visualize the extracted data respectively across the two antenna placement runs to determine the impact of the antenna location on performance. From a different perspective, the analysis was part of experimentation for the DRIVE-SAFE: Remote and Cooperative Autonomous Driving in Dynamic Environments using 5G/NextG Technology project on POWDER. Repeating the data analysis: We provide a POWDER profile that can automatically instantiate a small sandbox data lake (your own data pond) containing the data and including example scripts to extract and analyze data as described above. I.e., generating the CSV files associated with this dataset. The POWDER profile is available here. Running the POWDER profile: Pre-requisites: You need a POWDER user account and access to a project on POWDER that has permission use the CephFS storage cluster. You need to ensure your SSH key is added to POWDER; follow the POWDER manual under this sub-section. Steps: Instantiate a POWDER experiment using the POWDER data-pond profile from within a POWDER project that has permission to use the CephFS storage cluster. After it starts, refer to the "Instructions" section of the profile for steps on how to use the example scripts to (1) extract the analysis results shown here from a data pond containing raw logs, and (2) visualize the analysis results graph generated from the dataset.