Cooperative Spectrum Sensing With DeGAN and KANCNN for Nonorthogonal Multiple Access

Nonorthogonal multiple access (NOMA) technology achieves higher communication throughput compared with orthogonal multiple access, but it also introduces significant challenges for spectrum sensing (SS), particularly in accurately detecting channels occupied by multiple users in complex environments. To address these challenges and enhance SS capabilities in power-domain NOMA, this article proposes a novel deep learning-based algorithm that integrates a denoising generative adversarial network (DeGAN) for effective noise reduction. The DeGAN module employs a joint loss function to effectively suppress noise while preserving critical time–frequency features of the signals. Subsequently, the power spectral density of the denoised signals is extracted and utilized as a feature for sample classification through a model that integrates the Kolmogorov–Arnold networks–convolutional neural network (KANCNN). Comparative results demonstrate that the DeGAN–KANCNN algorithm surpasses other methods in detection accuracy and interference resistance in challenging environments.