Deep learning based hybrid LCCNN dual feature extraction framework for spectrum sensing in cognitive radio wireless sensor networks

Abstract Spectrum sensing, the ability to identify free bands in the radio spectrum, allows a cognitive radio system to establish a communication path free of interference, a key feature for efficient spectrum management in a dense environment. This work introduces a deep learning-based LCCNN (LMS-Covariance-CNN) dual feature extraction framework for improved spectrum sensing, employing the Least Mean Square equalizer, covariance analysis, and a CNN. The LMS equalizer mitigates channel-induced distortions and provides an equalized signal for better classification and feature extraction. The covariance matrix feature captures the statistical properties of the signal, or more precisely the correlations in the signal, for signal analysis, and the CNN learns temporal patterns from the covariance matrix features for better classification of signal presence in the channel. The feature extractors interact with each other and form a hybrid dual feature extraction framework based on temporal patterns and statistical correlations. The proposed LCCNN is robust and can be adaptable to noisy wireless environments. Performance outcomes show enhancement in probability of detection, false alarm rate, sensing error, and F1 score, showing the approach’s efficiency in sensing the spectrum in wireless sensor networks.