Observer Comparison Benchmark: ANO vs EKF vs HGO

This repository contains MATLAB codes and technical documents for a systematic comparison of three observers for nonlinear systems of the form

[ \dot{x} = A x + b\big(f(x)+g(x)u+d(t)\big), \qquad y = c^T x. ]

The observers studied in this project are:

• Adaptive Neural Observer (ANO)
• Extended Kalman Filter (EKF)
• High-Gain Observer (HGO)

Main result

An important outcome of the benchmark is that the ANO does not use the true nonlinear functions (f) and (g), while EKF and HGO are implemented with explicit nonlinear model knowledge. Despite this, ANO achieves comparable and often better performance in the tested scenarios.

Overall results on 63 generated scenarios:

• ANO: 62/63 full convergence (98.4%)
• EKF: 60/63 full convergence (95.2%)
• HGO: 46/63 full convergence (73.0%)

Under high noise:

• ANO: 31/32 full convergence (96.9%)
• EKF: 29/32 full convergence (90.6%)
• HGO: 19/32 full convergence (59.4%)

Repository structure

• src/ANO/ : implementation of the Adaptive Neural Observer
• src/EKF/ : implementation of the Extended Kalman Filter
• src/HGO/ : implementation of the High-Gain Observer
• tests/ : simulation and benchmark scripts
• docs/ : technical and comparative documents

Documents

• docs/Technical details of ANO.pdf
• docs/Theoretical foundation of observers.pdf
• docs/Comparative analysis of observers.pdf

Main scripts

• tests/run_observer.m
• tests/test1.m
• tests/test2.m

Author

Roozbeh Abolpour