Design of an Novel Spectrum Sensing Scheme Based on Long Short-Term Memory and Experimental Validation

TL;DR

This paper introduces a novel spectrum sensing method using LSTM deep learning, validated through real-world experiments, which improves detection accuracy under low SNR conditions compared to existing techniques.

Contribution

The paper presents a new LSTM-based spectrum sensing approach that learns features implicitly and is validated with real-world radio broadcast data, demonstrating superior performance.

Findings

Effective detection at low SNR levels

Improved classification accuracy over existing methods

Validated with real-world FM broadcast data

Abstract

Spectrum sensing allows cognitive radio systems to detect relevant signals in despite the presence of severe interference. Most of the existing spectrum sensing techniques use a particular signal-noise model with certain assumptions and derive certain detection performance. To deal with this uncertainty, learning based approaches are being adopted and more recently deep learning based tools have become popular. Here, we propose an approach of spectrum sensing which is based on long short term memory (LSTM) which is a critical element of deep learning networks (DLN). Use of LSTM facilitates implicit feature learning from spectrum data. The DLN is trained using several features and the performance of the proposed sensing technique is validated with the help of an empirical testbed setup using Adalm Pluto. The testbed is trained to acquire the primary signal of a real world radio broadcast taking place using FM. Experimental data show that even at low signal to noise ratio, our approach performs well in terms of detection and classification accuracies, as compared to current spectrum sensing methods.