Empirical Comparison of Full and Split gNB Architectures in 5G NTN Based on a Laboratory Testbed
TL;DR
This work profiles the CPU and memory footprints as well as the feeder link performance of the 3GPP-standardized split option 2 against the monolithic gNB, offering a perspective on functional network management, which leads to the identification of potential security enhancements, geographic-based Quality of Service (QoS) strategies, and handover optimizations.
Abstract
Ubiquitous connectivity remains a primary focus for 3GPP, driving the development of 5G Non-Terrestrial Networks (NTN) to integrate regenerative satellites directly into the network. To this end, 3GPP proposes two on-board architectural approaches: hosting a monolithic Next Generation Node B (gNB) or hosting only the Distributed Unit (DU), following the functional split architecture. The technical trade-off between both architectures has already been addressed in literature from a theoretical perspective, but this work goes a step further by providing an empirical analysis, specifically examining the 3GPP-induced feeder link overhead and the on-board resource utilization. By deploying an experimental testbed based on OpenAirInterface (OAI), this study profiles the CPU and memory footprints as well as the feeder link performance of the 3GPP-standardized split option 2 against the monolithic gNB. Results demonstrate that the split architecture yields asymmetric feeder link performance due to variations in the GPRS Tunnelling Protocol (GTP) header, reducing IP-level overhead by 11% in Mobile Originated (MO) traffic while increasing it by 16% in Mobile Terminated (MT) traffic. Furthermore, the CPU profiling reveals that the computational bottleneck resides in the Physical (PHY) layer, particularly due to the DFT/IDFT, LDPC decoder, and RX/TX pipeline processing. The memory profiling identifies buffering mechanisms of Radio Link Control (RLC) Data Radio Bearers (DRBs) and Media Access Control (MAC) Hybrid Automatic Repeat Request (HARQ) processes as the main consumers. Consequently, the split gNB architecture does not reduce satellite hardware utilization, breaking the preliminary perception of functional split benefits in embedded systems. However, this work expands the discussion by offering a perspective on functional network management, which leads to the identification of potential security enhancements, geographic-based Quality of Service (QoS) strategies, and handover optimizations. The empirical findings provide validated guidelines for satellite hardware dimensioning and network design of future 6G systems.