Hybrid CNN-ABiGRU-Based Spectrum Sensing in Cognitive Radio Using Cepstral Feature Fusion and Modulation Behavior

Cognitive Radio Networks (CRNs) enable Secondary Users (SUs) to take use of underutilized frequency bands that are assigned to a Primary User (PU), hence facilitating effective spectrum utilization. In this research a novel Deep learning (DL) based Cognitive Radio Spectrum Sensing (SS) using cepstral coefficients and modulation behavior has been proposed for accurate primary user detection. The proposed approach extracts multiple signal features, including energy, power spectrum, and cyclostationary characteristics, from observed radio frequency data. In-phase and quadrature (I/Q) components are further processed to analyze channel behavior using variance and logistic mapping, where stable channels are identified as suitable for SUs access. To enhance feature representation, Mel Frequency Cepstral Coefficients (MFCC) and Bark Frequency Cepstral Coefficients (BFCC) are extracted and combined using an early fusion strategy. Channel stability decisions are incorporated through late fusion. The fused features are then classified utilizing a hybrid Convolutional Neural Network (CNN) based Attention Bidirectional Gated Recurrent Unit (CNN-ABiGRU) model with an attention mechanism to determine PUs availability. Experimental results illustrates that the proposed model enhances an accuracy of 98.15%, specificity of 98.9%, precision of 97.2%, recall of 98.45%, and F1-score of 97.9%. The model outperforms existing models such as Long Short-Term Memory-Multilayer Perceptron (LSTM-MLP), Deep Hole, and multiple-antenna SS models based on convolutional neural networks (MASSnets) in terms of detection performance and robustness.