Modern Software Defined Radio (SDR) platforms are capable of multi-gigabit sampling rates, producing an unprecedented volume of digitised RF data and presenting a challenge to local real-time processing.  Applications such as cognitive radio and wideband spectrum monitoring drive the need for high–performance external processing and, in turn, high-speed data offloading.  This is far from a trivial problem, and with the absence of well-tested reference designs, implementing a low-latency, high-speed network interface remains a significant challenge. 

To confront this challenge, StrathSDR has developed the QSFP Data Offload Design.  This platform, built around the AMD RFSoC 4×2 platform, supports the direct digitisation of wideband RF signals with up to 2 GHz of instantaneous bandwidth.  The captured data is then offloaded via a 100 Gbit/s QSFP network link to a secondary processing resource, where GPU-based hardware acceleration enables real-time spectral analysis, presented at 60 frames per second.  Experimental validation demonstrates a sustained throughput of approximately 80 Gbit/s over the QSFP link, whilst using only modest FPGA resources. 

By enabling high-throughput, low-latency streaming of wideband RF data to external accelerators, the system addresses the challenge of local real-time processing limitations in modern SDR applications.