Spectrum Sensing Method Based on STFT-RADN in Cognitive Radio Networks

Area of Science:

• Wireless Communications
• Signal Processing
• Artificial Intelligence

Background:

• Traditional CNN-based spectrum sensing struggles with feature representation and extraction.
• Shallow networks limit the capabilities of existing algorithms in cognitive radio networks.
• Need for robust spectrum sensing under low signal-to-noise ratio (SNR) conditions.

Purpose of the Study:

• Propose a spectrum sensing algorithm overcoming limitations of traditional CNNs.
• Enhance feature extraction and utilization for improved cognitive radio network performance.
• Develop a robust method adaptable to various modulation schemes.

Main Methods:

• Utilizing short-time Fourier transform (STFT) to convert signals into time-frequency spectrograms.
• Employing a residual attention dense network (RADN) with residual in dense (RID) blocks and CBAM for deep feature extraction.
• Training the RADN as a classifier for spectrum sensing using time-frequency images.

Main Results:

• The STFT-RADN method significantly improves spectrum sensing performance, especially at low SNR.
• Demonstrated high detection probability and strong robustness across different modulation schemes.
• Outperformed traditional deep-learning-based spectrum sensing methods.

Conclusions:

• The proposed STFT-RADN algorithm offers a powerful solution for spectrum sensing in cognitive radio networks.
• This approach effectively addresses feature representation and extraction challenges.
• The method provides enhanced performance and robustness, particularly in challenging low SNR environments.