TRACER Software Switching at Waveform Timescales

The electromagnetic spectrum is becoming increasingly congested, requiring radio frequency (RF) processing systems to operate in both more complex and higher-bandwidth signal environments. Waveform-specific hardware accelerators cannot handle the diverse, computationally intensive processing required to respond to the environment, e.g., due to uncertainty about RF signal operating parameters, while also maintaining high device utilization and throughput. Much of the burden now falls on software to coordinate low-latency signal processing, decision making, and adaptive task scheduling, but traditional software stacks are orders of magnitude too slow to achieve this at waveform timescales. The TRACER project uses a hardware/software co-design process to develop a next-generation RF signal processing accelerator to operate in these challenging signal environments. In this paper, we describe the optimized TRACER software stack that enables RF signal processing applications to adapt their processing at waveform timescales. We demonstrate local RISC-V context switching times as low as 26 ns and accelerator switching times as low as 25 ns.