This repository contains a MATLAB implementation of a Distributed Distributionally Robust Kalman Filter framework for state estimation across networked systems. The framework supports multiple estimation methods and diffusion strategies to optimize performance in distributed sensor networks.

Overview

The DRKF framework implements various distributed estimation techniques for multi-agent systems where each agent has access to different sensor measurements. The implementation features:

• Distributed Kalman Filter (DKF) with multiple diffusion strategies
• Distributionally Robust Optimization (DRO) variants
• Network topology configuration
• Comprehensive Monte Carlo simulation capabilities
• Performance evaluation and visualization tools

Network Configuration

The default implementation uses a 4-node square network topology where:

• Each node connects to two adjacent nodes
• Nodes have different sensor configurations
• The system allows testing various estimation and fusion methods

Estimation Methods

The framework implements multiple estimation approaches:

1. Standard DKF: Classical Distributed Kalman Filter
2. DRO-based methods
   • KL Divergence-based robust estimation
   • Wasserstein distance-based robust estimation
   • Moment-based robust estimation

Diffusion Strategies

Three diffusion (information fusion) approaches are implemented:

1. No Diffusion: Each node relies solely on its local measurements
2. Average Diffusion: Simple averaging of state estimates from neighboring nodes
3. Covariance Intersection (CI): More sophisticated fusion that accounts for correlation between estimates

Key Files

• main.m: Main execution script that runs all experiments
• utils/: Directory containing utility functions:
  • getSys.m: System model initialization
  • getCorrData.m: Data generation with correlated noise
  • calculate_mse.m: Performance calculation functions
  • DRO.m: Implementation of distributionally robust optimization
  • initialize_nodes.m: Network configuration
  • plot_MSEs_new.m: Visualization tools