Sep 26 – 30, 2022
Capital Hilton
US/Eastern timezone
All GRCon talks are now available to watch at https://www.youtube.com/GNURadioProject

Inference as DSP: An Approach to Heterogeneous Compute-Enabled SDR Applications

Sep 29, 2022, 1:30 PM
15m
Presidential Ballroom (Capital Hilton)

Presidential Ballroom

Capital Hilton

Paper (with talk) High Performance SDR Applications Main Track

Speaker

Garrett Vanhoy (Peraton Labs)

Description

Constructing a framework for the development of SDR applications backed by heterogenous compute platforms has been the subject of significant development effort over a few decades, but no single framework has gained significant traction in the open-source SDR community. At the same time, machine-learning software has matured and made efficiently implementing tensor operations across CPU, GPU, and FPGA accessible even to novice developers. We present a novel GNU Radio out-of-tree (OOT) module that integrates with Triton Inference Server (TIS) -- effectively enabling highly-scalable SDR application development backed by heterogenous compute platforms. The new OOT brings exciting capabilities to the GR 3.x line of development which is currently available, but also presents the first significant use of the GR 4.x modular scheduler framework. Finally, we present a number of performance comparisons demonstrating the immediate benefit the new OOT brings to application developers.

Talk Length 15 Minutes
Acknowledge Acknowledge In-Person

Primary author

Garrett Vanhoy (Peraton Labs)

Presentation materials