Bayesian-Enhanced-AoA-Estimator

AoA estimator for passive UHF RFID based on Bayesian regression and classical antenna array signal processing. Combines physics-informed priors with Pyro-based uncertainty quantification.

📑 Table of Contents

• Bayesian-Enhanced-AoA-Estimator
  • 📑 Table of Contents
  • 🔍 Overview
  • 🧠 Bayesian Approach
  • 🚀 Getting Started
    • Quick Start
  • 📊 Dataset Structure
    • 📂 File Naming Convention
    • 📁 Directory Structure
  • 🧮 MATLAB Implementation
    • 📄 process_experimental_data.m
    • 📄 antenna_array_processing.m
  • 🐍 Python Implementation
    • 📄 bayesian_regression.py
    • 📄 beamforming.py
    • 📄 data_management.py
    • 📄 MUSIC.py
    • 📄 phase_difference.py
    • 📄 visualization.py
  • 📁 Repository Structure
    • /data
    • /figures
    • /MATLAB
    • /results
    • /src
  • 📊 Results and Performance
  • 📄 License

🔍 Overview

The Bayesian-Enhanced-AoA-Estimator provides a comprehensive framework for estimating the Angle of Arrival (AoA) in passive UHF RFID systems. This project combines:

1. Classical Antenna Array Processing: Implements traditional techniques like Phase-difference estimation, Delay-and-Sum beamforming, and MUSIC algorithm.

2. Bayesian Regression Approach: Leverages probabilistic programming with Pyro to incorporate physics-informed priors and estimate uncertainty.

3. Multi-frequency Fusion: Combines data from multiple frequencies to improve estimation accuracy and robustness.

4. Uncertainty Quantification: Provides confidence metrics for all estimates, essential for real-world deployment.

This approach significantly improves AoA estimation accuracy compared to classical methods alone, particularly in challenging low-SNR environments and multi-path scenarios typical in indoor RFID deployments.

🧠 Bayesian Approach

Our Bayesian approach offers several key advantages over traditional methods:

• Physics-Informed Priors: Incorporates domain knowledge from classical antenna array theory as priors, making the model robust even with limited data.

• Hierarchical Modeling: Employs a hierarchical Bayesian structure to model relationships between physical parameters and observations at multiple levels.

• Uncertainty Quantification: Provides full posterior distributions rather than point estimates, enabling confidence-aware decision making.

• Model Comparison: Systematically evaluates different prior structures (DS, Weighted, MUSIC, Phase) and feature configurations for optimal performance.

• Robustness to Noise: Handles measurement noise and environmental uncertainties through explicit probabilistic modeling.

The Bayesian model is implemented using Pyro, a flexible probabilistic programming framework built on PyTorch.

🚀 Getting Started

Quick Start

1. Clone the repository:

   git clone https://github.com/yourusername/Bayesian-Enhanced-AoA-Estimator.git
   cd Bayesian-Enhanced-AoA-Estimator
   

2. Run the main script:

   python main.py
   

📊 Dataset Structure

📂 File Naming Convention

All measurement files follow this standardized naming pattern:

File Naming Convention:
YYYYMMDD_FFF.F_D.DDD_L.LLL_W.WWW.csv

Explanation of Components:

• YYYYMMDD — Date of the experiment (for tracking only; does not affect the measurement).
• FFF.F — Operating frequency in MHz (e.g., 865.7 for 865.7 MHz). Used to compute λ.
• D.DDD — Vertical distance D in meters (e.g., 0.700 for 0.700 m).
• L.LLL — Inter-antenna spacing L in meters (e.g., 0.287 for 0.287 m).
• W.WWW — Horizontal offset W in meters (can be negative, zero, or positive).

---

📁 Directory Structure

The dataset is organized into a hierarchical directory structure as follows:

Distance 1/
├── Replica 1/
│   ├── Frequency 1/
│   ├── Frequency 2/
│   ├── Frequency 3/
│   └── Frequency 4/
├── Replica 2/
│   ├── Frequency 1/
│   ├── Frequency 2/
│   ├── Frequency 3/
│   └── Frequency 4/
└── Replica 3/
    ├── Frequency 1/
    ├── Frequency 2/
    ├── Frequency 3/
    └── Frequency 4/
Distance 2/
├── Replica 1/
│   ├── Frequency 1/
│   ├── Frequency 2/
│   ├── Frequency 3/
│   └── Frequency 4/
├── Replica 2/
│   ├── Frequency 1/
│   ├── Frequency 2/
│   ├── Frequency 3/
│   └── Frequency 4/
└── Replica 3/
    ├── Frequency 1/
    ├── Frequency 2/
    ├── Frequency 3/
    └── Frequency 4/
Distance 3/
├── Replica 1/
│   ├── Frequency 1/
│   ├── Frequency 2/
│   ├── Frequency 3/
│   └── Frequency 4/
├── Replica 2/
│   ├── Frequency 1/
│   ├── Frequency 2/
│   ├── Frequency 3/
│   └── Frequency 4/
└── Replica 3/
    ├── Frequency 1/
    ├── Frequency 2/
    ├── Frequency 3/
    └── Frequency 4/

Explanation:

• Distance X/: Represents different vertical distances D.
• Replica X/: Represents repeated measurements for the same distance.
• Frequency X/: Represents measurements taken at different operating frequencies.

🧮 MATLAB Implementation

The repository contains MATLAB scripts for processing RFID data and implementing various AoA estimation algorithms:

📄 process_experimental_data.m

A preprocessing script that:

• Batch processes RFID experiment CSV files from a COTS RFID system (Zebra FX7500, AN480 WB Antenna, Belt tag)
• Parses filenames to extract experimental parameters (frequency, distance, antenna spacing, etc.)
• Unwraps phase measurements and converts to radians
• Transforms RSSI values to linear power scale
• Creates complex phasors for antenna signals
• Organizes data into a structured MATLAB dataset (rfid_array_data.mat)

📄 antenna_array_processing.m

A comprehensive end-to-end RFID AoA estimation pipeline that:

• Implements multiple estimation methods:
  • Phase-difference estimation
  • Classical Delay-and-Sum beamforming (unweighted & RSSI-weighted)
  • MUSIC algorithm for high-resolution AoA
  • Multi-frequency fusion with confidence metrics
• Provides extensive visualization:
  • AoA vs. tag position plots
  • Spectral analysis and comparison
  • 3D beam pattern visualization
  • Heatmap representations
• Performs error analysis and method comparison
• Outputs organized figures and complete analysis reports

🐍 Python Implementation

The repository includes Python implementations that use Bayesian methods through Pyro:

📄 bayesian_regression.py

Core implementation of the Bayesian AoA estimator:

• Defines physics-informed observations based on antenna array geometry
• Implements probabilistic model for phase and RSSI observations
• Performs Bayesian inference using Pyro's SVI engine
• Provides posterior distributions for AoA estimates with uncertainty quantification
• Handles multi-frequency data fusion through hierarchical modeling

📄 beamforming.py

Provides functions to conduct classic antenna-array analysis of DS Beamforming and Weigthed DS Beamforming.

📄 data_management.py

Utility module for preprocessing and managing the dataset:

• Reads and parses CSV files from RFID experiments
• Converts raw measurements to complex phasors
• Handles data cleaning and outlier removal
• Provides data loaders compatible with PyTorch/Pyro

📄 MUSIC.py

Provides functions to conduct classic antenna-array analysis of the MUSIC algorithm.

📄 phase_difference.py

Provides functions to conduct classic antenna-array analysis of the phase difference analysis.

📄 visualization.py

Comprehensive visualization tools.

📁 Repository Structure

The repository is organized with the following key directories:

/data

Raw and processed datasets:

• /2025-07-09: Original CSV files from RFID experiments
• /testing: Data collected during environment and set up testing

/figures

Stores generated visualization outputs from the analysis:

• AoA estimation plots
• Spectral analysis visualizations
• 3D beam pattern representations
• Method comparison charts
• Error analysis visualizations

Example figures are included to demonstrate the expected output format.

/MATLAB

Contains all MATLAB implementation scripts:

• process_experimental_data.m: Preprocessing script for raw CSV data
• antenna_array_processing.m: Complete end-to-end AoA analysis pipeline

/results

Contains processed data and analysis results:

• rfid_array_data.mat: Preprocessed dataset ready for analysis
• complete_analysis.mat: Comprehensive results from all estimation methods
• ZIP files: Contains ZIP files of the full analysis pipeline.

Example result files are provided to illustrate the data structure.

/src

Contains all Python implementations:

• bayesian_regression.py: Core Bayesian estimation implementation
• beamforming.py: DS and Weighted DS Beamforming
• data_management.py: Dataset processing and management
• music.py: MUSIC algorithm
• phase_difference.py: Phase-difference methods
• visualization.py: Visualization tools

📊 Results and Performance

This section will include detailed performance metrics, comparisons, and visualizations of the Bayesian AoA estimator against classical methods.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.