DYNAMIC THRESHOLDING MECHANISM FOR CYCLOSTATIONARY SPECTRUM SENSING USING ARTIFICIAL NEURAL NETWORK UNDER TIME-VARIANT ENVIRONMENTAL CONDITIONS

Abstract

Reliable spectrum sensing is critical for cognitive radio networks to ensure efficient utilization of wireless resources and to protect primary users, particularly in dynamic maritime environments where channel conditions vary due to mobility, fading, and interference. Existing spectrum sensing techniques, including TFCFN, STFT-RADN, CNN-LSTM, and DBN-FOA, often suffer from limited adaptability to non-stationary channels, reduced detection accuracy, and suboptimal threshold selection, which can result in high false alarms or missed detections. To overcome these limitations, this work introduces a dynamic thresholding mechanism for cyclostationary spectrum sensing using artificial neural networks under time-variant environmental conditions. The proposed framework integrates Temporal Spectrum Variation Modeling (TSVM) via TSVANet, which combines ANN-based local spectral feature extraction with a Multi-Head Temporal Self-Attention (MHTSA) module to capture both local and long-term temporal dependencies, and Adaptive Learning Rate Optimization (ALRO) via Social Spider–Krill Hybrid (SKH), which dynamically tunes learning parameters for robust convergence under non-stationary data streams. Experimental evaluations demonstrate that the proposed method achieves 99.39% accuracy, 98.32% precision, and 99.09% sensitivity, outperforming existing approaches. The framework enhances spectrum sensing reliability, adaptability, and robustness, enabling efficient cognitive radio operation in complex maritime scenarios.