ML: Generalized Additive Model for Spectrum Sensing in Nakagami-m Fading Channel With Complex Generalized Gaussian Distribution Noise

Spectrum sensing in cognitive radio presents a challenge in overcoming the spectrum scarcity caused by the rapid growth of wireless devices. Deep learning (DL) and machine learning (ML) are trending approaches to enhance the spectrum sensing performance. This paper introduces the generalized additive model (GAM) as an ML-based spectrum sensing method for cognitive radio networks. In our proposed method, the received signals are transformed using fast Fourier transform (FFT) and the powers of both the real and imaginary components of the FFT signals are calculated separately to construct a dataset. In practical communication channels, spectrum sensing is a critical task for cognitive radio in the presence of non-Gaussian noise, such as noise with a heavy-tailed distribution. We consider the complex generalized Gaussian distribution (CGGD) as a noise model to evaluate the detection performance of the ML-based method in both Gaussian and non-Gaussian noise environments. Furthermore, generalized Gaussian distribution (GGD) noise is used to compare the results of the GAM algorithm with existing methods. As a channel model, the Nakagami-m fading distribution is used for its versatile characteristics, which enable it to model a wide range of fading channels by varying the value of a parameter m in its equation. The effectiveness and practical usefulness of the proposed method are validated through simulations. The simulation results exhibit higher performance compared to the state-of-the-art DL and existing conventional methods.