How does a Smith Predictor compensate for dead time?

Smith Predictor uses a process model to predict future behavior and subtract expected dead time response, allowing controller to act on predicted output before actual dead time elapsed. Structure: primary controller, process model without dead time, and dead time model. Model output (fast) minus dead-time-delayed model output creates prediction. Controller sees predicted response without dead time, enabling faster tuning. Limitations: sensitive to model accuracy (dead time and dynamics), amplifies model errors, and requires good model maintenance. Effective when dead time is dominant dynamic (theta/tau > 0.5).