Feedback and Stability in Real Systems

A bridge page showing how feedback turns measurements into actions, and why stability is the first nonnegotiable requirement for real systems.
Modified

April 26, 2026

Keywords

feedback, stability, tracking, closed-loop, control

1 Application Snapshot

Real systems rarely get one perfect command and then coast forever.

They are pushed by disturbances, limited by actuators, and measured through noisy or partial sensors.

That is why control almost always becomes a feedback problem:

measure what happened, compare it to what you wanted, and use the difference to choose the next action

The first question after that is not sophistication. It is stability.

2 Problem Setting

Suppose a system has state \(x\), output \(y\), and input \(u\).

In open loop, we choose the input ahead of time:

\[ u(t) = u_{\mathrm{plan}}(t). \]

In feedback, we let the input depend on what the system is doing now:

\[ u(t) = \pi(y(t), r(t)) \]

or, when the full state is available,

\[ u(t) = \pi(x(t), r(t)). \]

Here \(r\) is a desired reference, such as:

  • a target speed
  • a target height
  • a desired temperature
  • a desired path or trajectory

The difference between what we want and what we currently observe is often called the error:

\[ e(t) = r(t) - y(t). \]

Feedback uses that error to keep correcting the system as time passes.

3 Why This Math Appears

This language reuses several math layers already on the site:

  • ODEs and Dynamical Systems: stability asks what nearby trajectories do over time
  • Control and Dynamics: feedback changes the closed-loop dynamics, not just the command signal
  • Linear Algebra: in linear models, feedback often appears as \(u = -Kx + r\)
  • Signal Processing and Estimation: when the full state is not measured, feedback acts on estimates rather than the true state
  • Stochastic Control and Dynamic Programming: uncertainty and repeated decisions turn feedback into a sequential optimization object

So stability is not an isolated theorem topic. It is the first reality check on whether a feedback idea can be trusted at all.

4 Math Objects In Use

  • state \(x\)
  • output \(y\)
  • reference \(r\)
  • error \(e = r - y\)
  • control input \(u\)
  • feedback law \(\pi\)
  • closed-loop dynamics obtained after substituting the feedback law into the system

For a linear state-space model,

\[ \dot{x} = Ax + Bu, \qquad y = Cx, \]

full-state feedback of the form

\[ u = -Kx + r \]

changes the dynamics to

\[ \dot{x} = (A-BK)x + Br. \]

That new matrix is the closed-loop object whose behavior now matters.

5 A Small Worked Walkthrough

Consider a simple cruise-control model:

\[ m\dot{v} = -cv + ku + d(t), \]

where:

  • \(v\) is the car’s speed
  • \(u\) is throttle input
  • \(c\) models drag
  • \(d(t)\) is a disturbance such as a hill or wind

If we command a fixed throttle in open loop, then changes in slope or drag can easily push the speed away from what we wanted.

Now use feedback:

\[ u = K(v_{\mathrm{ref}} - v). \]

Then the input increases when the car is too slow and decreases when it is too fast.

Define the tracking error

\[ e = v_{\mathrm{ref}} - v. \]

Ignoring disturbance for a moment, the error dynamics become

\[ \dot{e} = -\frac{c + kK}{m} e. \]

That equation says something operational, not just symbolic:

  • if the coefficient is negative, the error shrinks
  • if it shrinks, the speed returns toward the target
  • if it grows, the controller is not stabilizing the behavior you want

This is exactly why stability is the first thing readers keep seeing in control papers and real systems notes.

6 Implementation or Computation Note

Three practical complications appear almost immediately:

  1. Delays and sampling Digital controllers act at discrete times, not continuously.

  2. Noise and partial sensing The controller may only see noisy measurements or a state estimate.

  3. Actuator limits Throttle, torque, force, and voltage usually saturate.

Use these pages as the strongest follow-on support:

7 Failure Modes

  • thinking feedback means “keep increasing the gain until it works”
  • assuming a stable open-loop plant automatically gives a stable closed-loop design
  • forgetting that delays and actuator saturation can break an otherwise clean feedback idea
  • treating tracking and stabilization as the same task without checking the reference signal
  • writing a controller around measurements without asking whether the key state is actually observed

8 Paper Bridge

  • 16.30 / Feedback Control Systems - First pass - official MIT course where feedback and closed-loop stability are the main operational language. Checked 2026-04-25.
  • EE363 / Linear Dynamical Systems - Paper bridge - a strong Stanford anchor once you want the linear-systems side of feedback to feel precise. Checked 2026-04-25.

9 Sources and Further Reading

Back to top