Deep Cooperative Spectrum Sensing Based on Residual Neural Network Using Feature Extraction and Random Forest...

Area of Science:

• Wireless Communications
• Signal Processing
• Machine Learning

Background:

• Increasing demand and user allocation policies have led to spectrum overload and underutilization.
• Cognitive radio (CR) offers dynamic spectrum allocation for unlicensed users.
• Cooperative spectrum sensing (CSS) shows promise for improving spectrum utilization efficiency.

Purpose of the Study:

• To propose a cooperative spectrum sensing approach using a residual neural network (ResNet) architecture.
• To achieve high accuracy in detecting licensed users, even under high noise power density.
• To reduce the number of cooperating unlicensed users required for effective spectrum sensing.

Main Methods:

• Feature extraction from sensing information of individual unlicensed users.
• Random forest classification to identify licensed user presence at each unlicensed user.
• Fusion of information from multiple unlicensed users at a central fusion center.
• Decision making using a ResNet model trained on fused sensing data.

Main Results:

• Achieved high accuracy in detecting licensed users, reaching 98% correct identification.
• Demonstrated effectiveness even with high noise power density (-134 dBm/Hz).
• Successful detection was achieved with the cooperation of only 10 unlicensed users.

Conclusions:

• The proposed cooperative spectrum sensing system effectively identifies licensed users in challenging noisy conditions.
• The combination of feature extraction, random forest, and ResNet offers a robust solution for dynamic spectrum access.
• This approach enhances spectrum efficiency by enabling reliable detection with fewer cooperating users.