Spectrum Sensing using Hybrid Model with Attention Mechanism for Low SNR Environments
Zakia Sultana, Aditi Roy, Suparna Sen · 2024
With the rising demand for wireless spectrum, effective spectrum sensing is essential for optimal resource allocation in cognitive radio systems. Detecting signals in low Signal-to-Noise Ratio (SNR) settings, such as -20 dB, is difficult owing to the high noise levels. This research offers a comparative analysis of two hybrid deep learning model that incorporates Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). Again, Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks are incorporated together with attention mechanisms, to increase signal detection accuracy and likelihood of detection at low SNRs. Experimental results show that the CNN-LSTM with attention mechanism outperforms standard models, with a 66.72% accuracy at SNR of -20dB and 61.9% detection probability for signals at -20 dB SNR. Conversely, the CNN-RNN outperforms standard models, with a 61.86% accuracy at SNR of -20dB and 64.68% detection probability for signals at -20 dB SNR. The application of attention processes allows the model to focus on the most important signal information, reducing the influence of noise.