When is Enough Enough? "Just Enough" Decision Making with Recurrent Neural Networks for Radio Frequency Machine Learning

TL;DR

This paper explores dynamic decision-making in recurrent neural networks for wireless signals, enabling efficient, real-time classification by using just enough input data based on signal complexity.

Contribution

It introduces and analyzes four approaches for 'just enough' decision metrics, enhancing efficiency in RF machine learning applications.

Findings

Recurrent neural networks can make reliable decisions with fewer input symbols.

Dynamic decision strategies improve efficiency in wireless communication tasks.

Four methods for 'just enough' decision making are evaluated for practical use.

Abstract

Prior work has demonstrated that recurrent neural network architectures show promising improvements over other machine learning architectures when processing temporally correlated inputs, such as wireless communication signals. Additionally, recurrent neural networks typically process data on a sequential basis, enabling the potential for near real-time results. In this work, we investigate the novel usage of "just enough" decision making metrics for making decisions during inference based on a variable number of input symbols. Since some signals are more complex than others, due to channel conditions, transmitter/receiver effects, etc., being able to dynamically utilize just enough of the received symbols to make a reliable decision allows for more efficient decision making in applications such as electronic warfare and dynamic spectrum sharing. To demonstrate the validity of this concept, four approaches to making "just enough" decisions are considered in this work and each are analyzed for their applicability to wireless communication machine learning applications.