Ultra-Reliable and Low-Latency Wireless Hierarchical Federated Learning: Performance Analysis

Area of Science:

• Wireless communication systems
• Machine learning security
• Information theory

Background:

• Wireless hierarchical federated learning (WHFL) accelerates model training but faces eavesdropping risks.
• Ultra-reliable and low-latency communication (URLLC) is crucial for 5G/6G, necessitating secure implementations.
• Physical layer security (PLS) is vital for protecting wireless communication.

Purpose of the Study:

• To propose a secure finite block-length (FBL) approach for multi-antenna URLLC-WHFL.
• To analyze the trade-offs between privacy, utility, and PLS in the proposed scheme.
• To address the eavesdropping vulnerability in WHFL within URLLC contexts.

Main Methods:

• Development of a secure finite block-length (FBL) approach for multi-antenna URLLC-WHFL.
• Characterization of the relationship between privacy, utility, and PLS.
• Simulations to evaluate scheme performance under varying eavesdropper channel state information (CSI) conditions.

Main Results:

• The proposed FBL approach achieves near-perfect secrecy.
• Learning performance remains unaffected by the security measures.
• The scheme demonstrates robustness against imperfect eavesdropper CSI.

Conclusions:

• A novel secure FBL method is presented for URLLC-WHFL.
• The approach effectively balances security and learning efficiency.
• This work offers a viable solution for secure URLLC-WHFL systems facing PLS challenges.