# Robust CF Filtering — IEEE L-CSS 2026

> **Robust Cognitive-Flexible Filtering under Noisy Innovation Scores**  
> T. Nuchkrua and S. Boonto  
> *IEEE Control Systems Letters (L-CSS)*, 2026  
> Status: Under Review

---

## Overview

This repository contains all Julia source code to reproduce the numerical
experiments in the paper. The proposed **margin-based switching rule**
suppresses spurious structure transitions under bounded score noise,
restoring all three stability properties of noiseless CF theory.

---

## Requirements

Julia 1.9+ with the following packages:

```julia
] add Distributions Statistics Random Printf
```

---

## Files

| File                             | Description                              |
|----------------------------------|------------------------------------------|
| `lcss_experiments_documented.jl` | Figures 2–3, Table I (Theorems 1–3)      |
| `fig3_scaling_documented.jl`     | Figure 4, scaling validation (Theorem 2) |

---

## Run

```julia
julia lcss_experiments_documented.jl   # Figs 2–3, Table I
julia fig_scaling_documented.jl        # Fig 4
```

Figures are saved to `figures/` automatically.

---

## Parameters

| Parameter | Value         | Description                  |
|-----------|---------------|------------------------------|
| `Np`      | 500           | Number of particles          |
| `M`       | 100           | Monte Carlo runs             |
| `T`       | 200           | Horizon length               |
| `α`       | 2.5           | Margin multiplier (δ = α·ε̄)  |
| `ε̄`       | 0.5, 1.5, 3.0 | Score noise bounds (Table I) |

---

## Key Result

| Method                    | E[N_T] (ε̄=0.5) | E[N_T] (ε̄=1.5) | E[N_T] (ε̄=3.0) |
|---------------------------|----------------|----------------|----------------|
| Exact CF (oracle)         | 0.3            | 0.3            | 0.3            |
| CF without margin (δ=0)   | 83.7           | 81.1           | 79.2           |
| **Robust CF (proposed)**  | **0.3**        | **1.3**        | **7.9**        |
| Thm. 2 bound              | 8.2            | 11.4           | 17.6           |

Robust CF empirical count stays well below Theorem 2 bound ✓

---

## Citation

```bibtex
@article{nuchkrua2026robustcf,
  author  = {Nuchkrua, T. and Boonto, S.},
  title   = {Robust Cognitive-Flexible Filtering under
             Noisy Innovation Scores},
  journal = {IEEE Control Systems Letters},
  year    = {2026},
  note    = {Under review}
}
```

---

## Webpage

[https://thanana.github.io/RobustCF.html)

---

© 2026 Thanana Nuchkrua | Control & Robotics Research Group