SISO PRB-to-Throughput System Identification and Model Validation in a Virtual 5G/O-RAN Testbed

This scientific technical report presents experimental system identification and retrospective model validation for downlink Physical Resource Block (PRB) control in the Sci_O-RAN virtual 5G/O-RAN testbed (Tb3). The study examines six independent SISO PRB transitions (T1–T6), comprising 5,540 downlink throughput observations. The analysis covers data quality, time-series characteristics, transient responses, static throughput mapping, and dynamic system identification. We investigate four candidate models (M0–M3), including first-order dynamics with event-referenced delay (M2) and a second-order dynamic extension (M3). We estimate model parameters using T1, T3, T5, and T6. Frozen models are subsequently evaluated on T2 and T4 without parameter refitting or model reselection. Although M3 provides a small improvement in training error, its mean retrospective holdout error reduction relative to M2 is approximately 0.149%. The results therefore do not establish a robust predictive advantage for the second-order model. The report discusses limitations concerning parameter identifiability, temporal resolution, autocorrelation, event-referenced timing, and the distinction between statistical fit and physical actuation delay. The archive includes the scientific report in PDF format, its LaTeX sources, supporting materials, and reproducibility information. The associated Experimental Dataset and Offline Research Software are provided as separate Zenodo records.