Deep Learning-Based Spectrum Sensing in Cognitive Radio Networks using Stacked LSTM: Performance Analysis of SNR and BER

Abstract

Rapid evolution of wireless communication technologies and emergence of 5G networks have addressed lack of spectrum resources, highlighting the need for innovative topologies to improve the spectrum utilization. Due to this reason, Cognitive Radio (CR) has evolved as an optimized solution to overcome these difficulties by allowing dynamic spectrum access, thereby reducing the under-usage of available spectrum. Considering this, precise and accurate spectrum sensing is essential for CR to identify unused spectrum and ensure minimal interference with licensed users. This paper introduces novel Stacked Long Short-Term Memory (LSTM) network-based spectrum sensing algorithm for achieving enhanced sensing accuracy in Cognitive Radio Networks (CRNs). Stacked LSTM model is developed to capture temporal dependencies in incoming signals, facilitating robust spectrum sensing even under dynamic environments. The effectiveness of the stacked LSTM model is determined using key metrics such as Signal-to-Noise Ratio (SNR) and Bit Error Rate (BER). To validate its adaptability across various noise situations, this model is trained on a diverse set of signals with differing SNR levels. Simulation results indicate that Stacked LSTM significantly improves spectrum sensing accuracy, specifically in low-SNR conditions, when compared to conventional energy detection approaches. In addition to this, BER analysis depicts that the proposed model attains high transmission reliability, even under difficult and challenging channel situations, by efficiently reducing the BER.