Over-the-Air-Computation Based Federated Learning

Established and simulated Over-the-Air computation based federated learning model using MATLAB/Simulink

Overview

Developed a novel approach to federated learning using Over-the-Air (OTA) computation, implementing both basic AM modulation and advanced OFDM-based communication architectures in MATLAB/Simulink.

Theoretical Framework

Theoretical framework of analog Over-the-Air computation based Federated Learning, showing the mathematical principles and system design.

Basic Implementation

Simulink implementation using basic AM modulation/demodulation with 4 local devices, demonstrating the fundamental OTA computation concept.

OFDM-based Architecture

Left: OFDM transmitter architecture for OTA computation. Right: OFDM receiver design with built-in synchronization and channel estimation.

Technical Details

  • Implemented two communication architectures:
    • Basic AM modulation/demodulation system
    • Advanced OFDM-based system with built-in synchronization
  • Developed federated learning algorithms:
    • Distributed model training
    • Parameter aggregation via OTA computation
    • Model convergence optimization
  • Conducted comprehensive simulations:
    • Channel estimation and compensation
    • Synchronization performance analysis
    • System throughput evaluation

Key Features

  • Analog Over-the-Air computation
  • Multi-device parameter aggregation
  • OFDM-based communication
  • Channel estimation and compensation
  • Distributed model training
  • Real-time parameter updates

Technical Stack

  • MATLAB/Simulink
  • Communication Systems
  • Signal Processing
  • Machine Learning
  • OFDM Technology
  • Channel Modeling
  • System Optimization

Project Advisor

Prof. Howard Yang