Existence and Design of Functional Observers for Time-Delay Systems with Delayed Output Measurements

This paper addresses the functional state estimation problem for linear time-delay systems with distinct state and measurement delays by proposing three observer structures, establishing algebraic existence conditions, and introducing a functional augmentation framework to facilitate the systematic design of observers with varying orders.

The Gist

Imagine you are trying to drive a car, but you have a very strange problem: you can only see the road through a rearview mirror that is slightly foggy, and the image in that mirror is delayed.

In the real world, this is exactly what happens in many complex systems like power grids, chemical plants, or internet-controlled robots. The "state" of the system (where the car is, how fast it's going) changes over time, but the sensors measuring it take time to send the data back. Sometimes, the system itself has a "lag" (like a heavy truck taking time to turn), and the measurement of that system has a different lag (like a slow internet connection).

This paper is about building a smart guesser (called a "Functional Observer") that can figure out exactly what the system is doing right now, even though the data it receives is old and delayed.

Here is the breakdown of the paper's big ideas using simple analogies:

1. The Problem: The "Mismatched Lag"

Usually, engineers assume that if the system is slow, the measurement is slow by the exact same amount. But in reality, the system might have a 2-second lag (the truck is heavy), while the camera sending the video has a 5-second lag (the internet is slow).

• The Old Way: Previous methods tried to build a guesser that was the exact same size as the number of things you wanted to know. If you wanted to know 1 thing (like speed), they built a 1-dimensional guesser.
• The Problem: Sometimes, because the lags don't match, a small guesser just can't do the math. It's like trying to solve a complex puzzle with only 3 pieces when you need 10. The math simply doesn't work out.

2. The Solution: Three Types of "Smart Guessers"

The authors propose three different designs for these guessers, depending on how messy the delays are. Think of them as different levels of "superpowers" for your guesser:

• Structure A (The Basic Guess): This is a simple guesser. It looks at the delayed data and tries to predict the current state. It works great if the delays are perfectly aligned or simple.
  • Analogy: Like a weather forecaster who looks at yesterday's temperature to guess today's. It works if the weather is predictable.
• Structure B (The Memory Keeper): If Structure A fails, we upgrade to Structure B. This guesser has an internal memory. It doesn't just look at the data; it remembers what the data looked like a few seconds ago and uses that history to correct its guess.
  • Analogy: Like a chess player who doesn't just look at the current board, but remembers the last 5 moves to predict the opponent's next move.
• Structure C (The Time Traveler): This is the most powerful version. It handles the most complex situations where the measurement delay is longer than the system's own delay. It keeps track of multiple "time layers" (what happened 2 seconds ago, 5 seconds ago, etc.) to reconstruct the present.
  • Analogy: Like a detective who interviews witnesses from different time zones to reconstruct a crime scene that happened hours ago.

3. The Secret Weapon: "Generalized Functionals"

This is the paper's most creative idea.

Usually, if a guesser fails, you have to give up or build a massive, clumsy machine to fix it. The authors say, "No, let's change the question."

Instead of just asking, "What is the speed right now?" (which might be impossible to guess with a small device), they say, "Let's guess the speed plus the speed from 2 seconds ago."

• The Metaphor: Imagine you are trying to guess the exact position of a runner. It's hard. But if you guess "The runner's position and where they were 2 seconds ago," you actually have more information to work with. This extra information gives the math the "wiggle room" it needs to solve the puzzle.
• By guessing a slightly larger, more complex picture (an "augmented" state), the math suddenly becomes solvable, and you can still extract the exact speed you wanted at the end.

4. The "Recipe" (Algebraic Conditions)

The paper provides a strict set of rules (mathematical recipes) to tell engineers:

1. Check the ingredients: Does your system have the right properties to allow a guesser? (The "Rank Conditions").
2. Pick the right tool: Should you use Structure A, B, or C?
3. Build it: If the rules pass, the paper gives you the exact numbers (matrices) to plug into your computer to build the observer.

Why This Matters

In the real world, we can't always wait for perfect data.

• Networked Control Systems: When controlling a robot on Mars, the signal takes minutes to arrive.
• Power Grids: Sensors are far apart, and data travels at different speeds.
• Biological Systems: Drug effects in the body take time to show up in blood tests.

This paper gives engineers a toolkit to build stable, reliable controllers for these messy, delayed systems. It proves that even if your data is old and your system is slow, you can still build a "smart guesser" that knows exactly what is happening right now, keeping everything safe and stable.

In short: The paper teaches us how to build a time-traveling detective that can solve a mystery even when the clues arrive late and out of order, by using a little bit of extra memory and a clever change of perspective.

Technical Summary

Here is a detailed technical summary of the paper "Existence and Design of Functional Observers for Time-Delay Systems with Delayed Output Measurements."

1. Problem Statement

The paper addresses the problem of functional state estimation for linear time-delay systems. Specifically, it tackles a scenario often found in networked control systems where two distinct types of delays exist and are not aligned:

• State Delay ($\tau$): The delay affecting the system's internal state evolution ($x(t-\tau)$).
• Measurement Delay ($h$): The delay affecting the availability of output measurements ($y(t)$).

Key Challenge: Existing literature typically assumes instantaneous output availability or assumes the measurement delay matches the state delay ($h = \tau$). In practice, sensing and communication latencies often result in $h \neq \tau$. The goal is to estimate a desired linear functional $z(t) = Fx(t)$ (where $F$ is a full row-rank matrix) using only delayed output measurements, without reconstructing the full state vector.

2. Methodology

The authors propose a unified framework that distinguishes between the two delay cases ($h = \tau$ and $h > \tau$) and introduces three distinct functional observer structures to accommodate different delay configurations and observer orders.

A. System Modeling and Augmentation

The paper highlights that time-delay systems are inherently infinite-dimensional. To handle this, the authors utilize an augmented state space approach:

• If $h \leq \tau$, the augmented state is $\xi(t) = [x(t)^T, x(t-\tau)^T]^T \in \mathbb{R}^{2n}$.
• If $h > \tau$, the augmented state is $\xi(t) = [x(t)^T, x(t-\tau)^T, x(t-h)^T]^T \in \mathbb{R}^{3n}$.

This leads to the concept of Generalized Functionals, defined over these augmented vectors, allowing for greater design flexibility.

B. Proposed Observer Structures

Three structures are introduced, increasing in complexity and flexibility:

1. Structure-A (Delay-Free Internal Dynamics):
   • Uses delayed outputs and inputs but has no internal delay dynamics in the observer state ($w(t)$).
   • Suitable for minimal-order designs when specific rank conditions are met.
2. Structure-B (Internal Delay Dynamics):
   • Incorporates internal delay terms ($w(t-\tau)$) in the observer dynamics.
   • Allows for stabilization even when Structure-A fails (e.g., when the required observer matrix $N$ is not Hurwitz).
3. Structure-C (Generalized with Multiple Delays):
   • Designed specifically for the case $h > \tau$.
   • Incorporates multiple internal delay channels ($w(t-\tau), w(t-h)$) and delayed output terms ($y(t-\tau), y(t-h), y(t-2\tau)$, etc.).
   • Capable of estimating generalized functionals involving $x(t), x(t-\tau),$ and $x(t-h)$.

C. Design and Synthesis

The design process relies on:

• Algebraic Conditions: Deriving necessary and sufficient conditions for the existence of observers by decoupling the estimation error dynamics from the plant state. This involves solving linear matrix equations of the form $X\Theta = \Upsilon$.
• Functional Augmentation: If a minimal-order observer (order $r = \text{rank}(F)$) does not exist, the authors propose augmenting the functional to be estimated (adding extra functionals $R x(t)$ or delayed functionals $F_d x(t-\tau)$) to increase the observer order and satisfy existence conditions.
• Stability Analysis:
  • For Structure-A, stability requires the matrix $N$ to be Hurwitz.
  • For Structure-B and C, stability involves delay-dependent systems. The authors provide Linear Matrix Inequality (LMI) conditions (based on Lyapunov-Krasovskii functionals) to ensure asymptotic stability of the error dynamics.

3. Key Contributions

1. Distinction of Delays: Explicitly modeling and solving the functional observer problem where state delay ($\tau$) and measurement delay ($h$) are distinct and potentially mismatched.
2. Three-Structure Framework: Proposing a hierarchy of observer structures (A, B, and C) that allows for systematic design based on the specific delay configuration and the solvability of algebraic conditions.
3. Generalized Functionals & Augmentation: Introducing the concept of generalized functionals over augmented state vectors. This allows the design of higher-order observers when minimal-order solutions do not exist, providing a constructive method to expand the functional subspace.
4. Constructive Synthesis Procedures: Providing explicit algebraic conditions (rank tests) and LMI-based conditions for the existence and synthesis of these observers, including parameterization of the free matrices (e.g., $Z$) used to stabilize the system.

4. Results

The paper validates the theory through numerical examples:

• Case $h = \tau$: Demonstrated that when minimal-order Structure-A fails (due to rank conditions or instability of $N$), higher-order Structure-A (via augmentation) or Structure-B (via internal delays) can successfully estimate the functional.
• Case $h > \tau$: Showed that Structure-B might fail to stabilize the error dynamics for large delays, but Structure-C (using the full augmented state and multiple delays) can successfully achieve asymptotic stability.
• Stability Verification: The examples confirm that the proposed LMI conditions are feasible and result in asymptotically stable error dynamics, even for systems with unstable open-loop eigenvalues.

5. Significance

• Practical Applicability: The work bridges a critical gap between theoretical observer design and practical networked control systems, where communication delays often differ from physical process delays.
• Flexibility: By allowing for variable observer orders and generalized functionals, the framework overcomes the limitations of fixed-order designs that often fail in complex delay scenarios.
• Systematic Approach: The paper moves beyond trial-and-error by providing a rigorous, step-by-step algebraic and LMI-based procedure to determine if an observer exists and how to construct it.
• Foundation for Future Work: The methodology sets the stage for extending functional estimation to time-varying delays, uncertain parameters, and nonlinear systems, as well as integrating these observers into output-feedback stabilization loops for networked control.