cuSat: An Open GPU-Based DVB-S2X Feeder Link for Cellular-Satellite Convergence
Abstract
Cellular networks are expanding beyond terrestrial infrastructure, with Non-Terrestrial Networks (NTNs) emerging as a way to extend coverage, backhaul remote sites, and host parts of the Radio Access Network (RAN) stack in space. Yet most experimental platforms still treat the satellite feeder as an idealized pipe. This hides the physical effects that influence how well a RAN can actually run over a satellite link: amplifier-limited operation, waveform overhead, synchronization, Forward Error Correction (FEC) compute, path-loss dynamics, and dynamic link adaptation. Existing open cellular testbeds expose the RAN but abstract the feeder, while satellite emulators model the feeder without carrying a real waveform or RAN traffic. We present cuSat, an open-source framework for satellite-RAN converged research comprising a real-time DVB-S2X feeder Physical Layer (PHY) that leverages Graphics Processing Unit (GPU) compute to make the feeder software-defined, fast enough for production-scale transponder bandwidths, and ready for AI-based enhancements. cuSat drives the DVB-S2X waveform on Software-Defined Radios (SDRs), and is capable of coexisting with a RAN base station on the same radio. cuSat can also insert emulated orbital channel dynamics before the receiver for experimentation. cuSat sustains bidirectional operation over 36, 54, and 72 MHz transponder configurations using one GH200 2g.24gb MIG partition per endpoint, i.e., 24% of the GPU computational resources. Across 14 representative Modulation And Coding Schemes (MODCODs), measured application throughput ranges from 14.3 to 313 Mbps and stays within 88 to 97% of the idealized DVB-S2X information rate before pilot and packet overheads. With Adaptive Coding And Modulation (ACM) enabled, cuSat exposes how path loss, throughput, and Round-Trip Time (RTT) interact across LEO-, MEO-, and GEO feeder links. We then demonstrate how cuSat can be used as the GEO backhaul of an Over-the-Air (OTA) 5G cell, preserving throughput relative to the ethernet-based baseline while adding propagation delay, and sharing the radio and compute resources between backhaul and RAN. Finally, we lay research directions that cuSat enables for the research community.