Deep learning inference in GNU Radio with ONNX

Abstract

This paper introduces gr-dnn, an open source GNU Radio Out Of Tree (OOT) block capable of running deep learning inference inside GNU Radio flow graphs. This module integrates a deep learning inference engine from the Open Neural Network Exchange (ONNX) project. Thanks to the interoperability with most of the major deep learning frameworks, it does not impose any restriction on the tool used by the model designer. As an example, we demonstrate here its functionalities running a simple deep learning inference model on raw radio samples acquired with a PlutoSDR.