Federated Learning for 5G Radio Spectrum Sensing.

Area of Science:

• Wireless communication
• Signal processing
• Machine learning

Background:

• Spectrum sensing (SS) is crucial for dynamic spectrum access by secondary users (SUs) to avoid interfering with primary users (PUs).
• Deep learning (DL) offers intelligent SS but requires extensive data and computational resources, posing challenges for SUs.
• Existing DL models for SS are often specialized for specific radio environments, limiting their general applicability.

Purpose of the Study:

• To propose a novel Federated Learning (FL) approach for spectrum sensing to overcome data and training limitations for SUs.
• To enhance the efficiency and accessibility of spectrum sensing for unlicensed users in dynamic radio environments.
• To enable collaborative and distributed training of DL models for SS without centralizing sensitive data.

Main Methods:

• Implemented a Federated Learning (FL) algorithm to distribute data collection and model training across multiple devices.
• Categorized FL devices into groups based on their mean Signal-to-Noise ratio (SNR) for targeted model training.
• Developed an iterative process to create a common DL model for each SNR-based group.

Main Results:

• The FL algorithm achieved spectrum detection accuracy comparable to traditional, specialized Deep Learning (DL) models like Convolutional Neural Networks (CNNs).
• The proposed method successfully simplified the spectrum sensing process within the network.
• Grouped training based on SNR demonstrated effective adaptation of DL models to varying radio conditions.

Conclusions:

• Federated Learning (FL) provides an effective and scalable solution for spectrum sensing (SS) challenges faced by secondary users (SUs).
• The SNR-based grouping in FL enhances the performance and relevance of DL models for spectrum detection.
• This approach facilitates intelligent spectrum sharing by making advanced SS techniques more accessible to SUs.