Computing Library › Control Theory
Control Theory

Discrete-Time Control

Digital controllers act at discrete sample instants, requiring difference equations, the z-transform, and attention to sampling effects.

Control by a Computer

Nearly all controllers today run on digital computers, which read sensors, compute, and update actuators at discrete instants rather than continuously. Discrete-time control is the theory of such sampled systems, where signals exist only at sample times and dynamics are described by difference equations instead of differential equations.

The sampling process

Kronos motion — control room

The z-domain

Just as the Laplace transform serves continuous systems, the z-transform serves discrete ones. It turns difference equations into algebraic equations in the variable z, yielding discrete transfer functions and pole-zero descriptions. The stability region changes accordingly: discrete poles must lie inside the unit circle rather than in the left half-plane.

Discretization

A controller may be designed in continuous time and then converted to discrete form, using methods such as the zero-order-hold equivalent, the bilinear (Tustin) transform, or matched pole-zero mapping. Alternatively the plant is discretized first and the controller designed directly in the z-domain. Each path has trade-offs in accuracy and fidelity of the frequency response.

Sampling effects

Sampling introduces effects absent in continuous control. Too slow a sample rate degrades performance and can destabilize a loop, since the controller sees the plant only intermittently and reacts with delay. The zero-order hold itself adds phase lag of roughly half a sample period. And frequencies above half the sample rate alias into lower ones, corrupting measurements unless removed by an anti-aliasing filter.

Rule of thumb

A common guideline is to sample at least ten to twenty times faster than the closed-loop bandwidth, giving the digital controller enough resolution to behave like its continuous counterpart. Fast loops, such as real-time plasma position control, demand correspondingly high sample rates and low-latency computation.