LECTURE ROOM/35 — Sampling and command paths
CHAPTER 35 / Sampling and command paths

From the left half-plane to the unit circle

Keep the same continuous mode and change the sampling period. Does a smaller discrete pole mean faster motion in seconds?

01 / THE INTUITION

First, picture it

A continuous exponential mode is multiplied by exp(sT) every T seconds. Calling this multiplier z gives x[k+1]=zx[k] for the complex mode. The real trace here is exp(σt)cos(ωt); the two crosses are a conjugate pair. We plot its exact samples, not a numerical approximation or a redesigned controller. Start at σ=−0.5, ω=4 rad/s and T=0.2 s.

02 / THE IDEA

What the model says

Magnitude is exp(σT). Negative σ gives a factor below one, so repeated multiplication decays. Positive σ gives growth. At σ=0, the factor is exactly one: this isolated mode does not decay and is not asymptotically stable. For a full finite-dimensional LTI realization every eigenvalue must lie strictly inside the unit circle for asymptotic stability. A repeated defective boundary pole can grow even without a radius above one.

RELATIONs = σ ± jω → z = exp(sT), |z| = exp(σT), x(t) = exp(σt) cos(ωt)

Angles differing by 2π give the same z, so frequencies separated by 2π/T rad/s cannot be distinguished by these samples alone. Exact modal sampling preserves decay; implementing a feedback controller with a hold, computational delay or an approximate integrator is another problem. Do not infer that any slow sampled controller is safe from this picture. The bench has no actuator or feedback loop.

01 / STEP BY STEP

One mode, two clocks

A continuous exponential mode is multiplied by exp(sT) every T seconds. Calling this multiplier z gives x[k+1]=zx[k] for the complex mode. The real trace here is exp(σt)cos(ωt); the two crosses are a conjugate pair. We plot its exact samples, not a numerical approximation or a redesigned controller. Start at σ=−0.5, ω=4 rad/s and T=0.2 s.

02 / STEP BY STEP

Why the boundary is a circle

Magnitude is exp(σT). Negative σ gives a factor below one, so repeated multiplication decays. Positive σ gives growth. At σ=0, the factor is exactly one: this isolated mode does not decay and is not asymptotically stable. For a full finite-dimensional LTI realization every eigenvalue must lie strictly inside the unit circle for asymptotic stability. A repeated defective boundary pole can grow even without a radius above one.

03 / STEP BY STEP

Compare seconds, not only samples

Record a prediction that |z| decreases, then change T from 0.2 to 0.5 s. The radius changes from about 0.905 to 0.779. Yet the decay envelope after one second is exp(−0.5) in both recordings: the physical mode has not changed. Larger intervals contain more decay per step but fewer steps per second. Pin both traces and inspect the same physical time.

04 / STEP BY STEP

A map is not a digital controller design

Angles differing by 2π give the same z, so frequencies separated by 2π/T rad/s cannot be distinguished by these samples alone. Exact modal sampling preserves decay; implementing a feedback controller with a hold, computational delay or an approximate integrator is another problem. Do not infer that any slow sampled controller is safe from this picture. The bench has no actuator or feedback loop.

03 / IN PRACTICE

Make it concrete

Keep the same continuous mode and change the sampling period. Does a smaller discrete pole mean faster motion in seconds?

Open this experiment
BE CAREFUL

Angles differing by 2π give the same z, so frequencies separated by 2π/T rad/s cannot be distinguished by these samples alone. Exact modal sampling preserves decay; implementing a feedback controller with a hold, computational delay or an approximate integrator is another problem. Do not infer that any slow sampled controller is safe from this picture. The bench has no actuator or feedback loop.

Check your understanding+

Keep the same continuous mode and change the sampling period. Does a smaller discrete pole mean faster motion in seconds?

Record a prediction that |z| decreases, then change T from 0.2 to 0.5 s. The radius changes from about 0.905 to 0.779. Yet the decay envelope after one second is exp(−0.5) in both recordings: the physical mode has not changed. Larger intervals contain more decay per step but fewer steps per second. Pin both traces and inspect the same physical time.