Open-Source VoWiFi Over 4G/5G CUPS Architecture: Implementation, QoS Analysis, Perceptual Quality Assessment, and Resource Characterization

Abstract

Voice over WiFi (VoWiFi) has emerged as a strategic solution for mobile network operators to extend indoor coverage and reduce infrastructure costs. Despite this, the research community lacks accessible, open-source testbeds that accurately replicate production-grade VoWiFi deployments over 4G/5G Control and User Plane Separation (CUPS) architectures. This paper presents the implementation and comprehensive experimental evaluation of an open-source VoWiFi testbed integrating containerized Open5GS network functions, WiFi Access Point infrastructure, and IP Multimedia Subsystem (IMS) components. The implementation demonstrates one of the first open-source CUPS deployments supporting VoWiFi over untrusted WiFi access via an evolved Packet Data Gateway (ePDG), including configuration of the Packet Forwarding Control Protocol (PFCP) interfaces (Sxa, Sxb) and SWu, SWx, and S2b connectivity. A multi-dimensional measurement framework is developed encompassing: (i) network-level QoS metrics (jitter, Round-Trip-Time-RTT, throughput), (ii) perceptual quality scores (R-factor, Mean Opinion Score-MOS) derived via ITU-T G.107 E-Model, and (iii) computational resource characterization (Central Processing Unit - CPU utilization per container across operational phases). Experimental evaluation across progressive load conditions baseline, single User Equipment (UE) attachment, dual UE attachment, and active VoWiFi calls, reveals that the open-source implementation achieves jitter performance comparable to high performance commercial reference equipment (mean jitter: 2.54–4.14 ms vs. 2.66 ms Anritsu; 94.3% below 20 ms QoS threshold), RTT within ITU-T G.114 delay budgets (mean propagation delay: 46–50 ms), and toll-quality voice scores (MOS: 4.36–4.42). Resource characterization uncovers a three-tier CPU architecture: an infrastructure tier (1.7% baseline), an event-driven control plane (<0.15%), and a traffic-proportional data plane (30-60x scaling during active calls), with the total system footprint remaining below 5% even under full VoWiFi load. These findings validate that containerized open-source 5G CUPS implementations can achieve perceptual voice quality equivalent to that of proprietary systems while maintaining efficient resource utiliz ation, thereby establishing a reproducible platform for VoWiFi research and capacity planning.