<![CDATA[Private 4G LTE networks built on open-source platforms provide cost-effective localized connectivity. However, executing baseband Digital Signal Processing (DSP) imposes a massive computational burden on general-purpose CPUs. This study evaluates the correlation between DSP load and end-to-end Quality of Service (QoS) in a USRP B210-based Private 4G LTE prototype using srsRAN and Open5GS. To prevent CPU saturation, Advanced Vector Extensions 2 (AVX2) SIMD instructions and CPU core affinity were applied. The experimental design evaluated the system using two concurrent User Equipment (UEs) under Line-of-Sight (LOS) outdoor conditions. Compared to a non-optimized baseline, the AVX2 optimization successfully stabilized the host processor at a 60.8% peak load and achieved a downlink throughput of 35.3 Mbps without radio underflows. However, multiplexing complex baseband signals under heavy concurrent traffic severely congested the OS-level queues. This caused severe bufferbloat, escalating the average end-to-end latency to 2021 ms compared to its idle state. These findings demonstrate that while AVX2 optimization resolves the CPU bottleneck, software-induced bufferbloat remains the primary limitation in SDR deployments.]]>
Copyrights © 2026
