Performance Analysis of Wavelet-Denoising Convolutional Neural Networks for Spectrum Sensing under Diverse Modulation Schemes

Abstract

Spectrum sensing is a critical function in cognitive radio (CR) systems, enabling dynamic spectrum access while minimizing interference with primary users (PUs). This paper presents a performance analysis of a Wavelet-Denoising Convolutional Neural Network (WD-CNN) for spectrum sensing under diverse modulation schemes and low signal-to-noise ratio (SNR) conditions. The proposed approach integrates wavelet denoising with spectrogram-based deep learning to enhance the discriminative representation of received signals. Extensive simulations were conducted for BPSK, QPSK, and 16-QAM signals across low-SNR scenarios ranging from −5 dB to −20 dB. Comparative evaluation with a pre-trained AlexNet demonstrates that WD-CNN consistently achieves higher classification accuracy and reduced spectrum sensing latency, particularly under severe noise conditions and for higher-order modulation schemes. The results confirm that the proposed WD-CNN framework provides a robust and computationally efficient solution for real-time spectrum sensing in CR networks.