Time-Delay Systems
Pure time delays add frequency-dependent phase lag that limits achievable bandwidth and can destabilize otherwise sound loops.
When the Effect Lags the Cause
Many processes respond only after a fixed delay: transport of material along a pipe, communication latency, or computation time all insert dead time between a command and its effect. A time delay of T seconds multiplies a signal's transform by e^(-s*T), a term that is not a rational polynomial and that behaves very differently from ordinary lags.
Why delay is hard
- A pure delay has unity magnitude at every frequency but adds phase lag that grows without bound as frequency rises.
- That accumulating phase lag eats into phase margin, so a loop that would be stable without delay can oscillate with it.
- Delay places a hard ceiling on achievable closed-loop bandwidth: you cannot react faster than the delay allows.
- Polynomial stability tests such as Routh-Hurwitz do not apply directly, so frequency-domain or specialized methods are used.
Effect on control
Because feedback cannot correct an error until the delayed effect appears, a controller acting on delayed information must be cautious. Aggressive gains that ignore the delay cause overshoot and instability. The general rule is that the loop bandwidth must stay well below the reciprocal of the delay, which fundamentally limits performance for long-dead-time processes.
The Smith predictor
A classic remedy is the Smith predictor, which uses a model of both the plant and its delay to predict what the output will be once the delay has elapsed, and controls against that prediction. If the model is accurate, the delay is effectively removed from the feedback loop, allowing higher gains. Its weakness is sensitivity to model error, especially error in the delay estimate itself.
Other approaches
Delay can be approximated by a rational Pade approximation to bring it into the polynomial framework, at the cost of introducing a non-minimum-phase zero. Model-predictive control handles delay naturally, since its internal model simply includes the dead time. Robust methods design directly for the delay's phase penalty.
Dead time is one of the most common reasons a straightforward PID loop underperforms, and recognizing it early guides the choice of a delay-aware controller.