6G-SANDBOX

The REPLICA project successfully collected and analyzed a comprehensive dataset from physical 5G SA experiments and complementary simulation activities. The comparison between real-world and simulated throughput results demonstrated that pure 3GPP-based simulations provided the closest match to experimental observations. This was largely due to the inclusion of features such as Time Division Duplex (TDD) partitioning, which were absent from the other models.

The machine learning–based path loss model showed potential for adaptive prediction across different scenarios but still requires refinement to improve accuracy under more variable conditions. Meanwhile, the trace-based simulation approach enabled realistic replay of test conditions but was inherently limited to specific test traces.

Across all configurations, higher variance in throughput was consistently observed at greater distances or under attenuated conditions, validating the strong impact of environmental factors on 5G NR performance, as expected. These results provide valuable insight for refining simulation models and enhancing the design and operational robustness of testbeds like 6G-SANDBOX, particularly in the context of trace-driven and reproducible experimentation.