Dual attention based CNN for multi-user spectrum sensing in cognitive radio systems

In cognitive radio networks, cooperative spectrum sensing (CSS) is a key approach for accurately detecting primary user activity in low-signal-to-noise-ratio scenarios. However, noise instability, multipath fading, and insufficient feature extraction often degrade performance in traditional sensing systems. The investigation proposes a dual-attention convolutional neural network (DA-CNN) for multi-user CSS based on in-phase and quadrature interpretations of the signals to address these obstacles. The proposed structure enables discriminative feature learning from noisy spectrum observations by simultaneously merging spatial attention and channel attention methods. Additionally, a feature-level cooperative fusion approach is proposed for effectively combining data from several secondary users, improving sensing accuracy in dynamic wireless channel conditions. A binary hypothesis approach was utilised on the RadioML2018.01A dataset, where the occupied channel state ( H 1 ) was expressed by the actual signal samples and complex-additive white Gaussian noise (AWGN) expressed as the idle channel state ( H 0 ) . AWGN is employed for modelling a realistic cooperative sensing environment over a broad signal-to-noise ratio (SNR) range. The proposed DA-CNN significantly outperforms conventional CNN and current attention-based techniques, especially in low-SNR situations, based on experimental results assessed using probability of detection, probability of false alarm, and receiver operating characteristic analysis. The following increases detection accuracy, robustness, and computational efficiency for intelligent cognitive radio-based applications.