Enhanced sensing performance through the integration of denoising autoencoder and ensembling techniques

Area of Science:

• Wireless Communication
• Signal Processing
• Cybersecurity

Background:

• Cognitive radio networks (CRNs) rely on collaborative spectrum sensing for efficient spectrum use.
• Malicious users, or False Sensing Users (FSUs), compromise sensing reliability by sending deceptive reports.
• FSUs exploit spectrum resources by disrupting the fusion center's (FC) decision-making process.

Purpose of the Study:

• To develop a robust method for detecting and mitigating the impact of FSUs in CRNs.
• To enhance the reliability of collaborative spectrum sensing against sophisticated attacks.
• To evaluate the performance of various machine learning classifiers in conjunction with a denoising autoencoder for improved spectrum sensing.

Main Methods:

• A denoising autoencoder (DAE) was employed to preprocess sensing data, reducing noise and mitigating abnormal reports.
• Multiple machine learning classifiers, including decision trees, k-nearest neighbor, neural networks, ensemble classification, Gaussian naive Bayes, and random forest, were assessed.
• The DAE-processed data was fed into these classifiers to estimate channel availability and make global decisions.

Main Results:

• The integration of DAE with ensemble classification (EC) demonstrated superior performance.
• High accuracy, F1 score, and Matthew's Correlation Coefficient (MCC) were achieved using the DAE-EC approach.
• The proposed method effectively minimized sensing errors, leading to reliable global decisions at the FC.

Conclusions:

• The DAE-EC framework provides a reliable solution for enhancing spectrum sensing in CRNs.
• This approach effectively counters the disruptive effects of False Sensing Users.
• Future work should include real-world radio frequency validation to complement synthetic data analysis.