FREQUENCY-DOMAIN LIMITS 8 MIN READ

Dead time: the limit no gain can buy back

Delay leaves the gain alone and takes phase in proportion to frequency. On the lab's motor 100 ms costs 19.5° at crossover; only a slower loop wins it back.

Dead time does not change how strongly the plant responds, only when. On a Bode plot that means an unchanged magnitude and a phase lag that grows without limit as frequency rises. Most other shortfalls in a loop can be traded against gain or compensation; this one puts a ceiling on how fast any controller can make the loop.

01 / FIELD NOTE

Phase in proportion to frequency

A delay of L seconds multiplies the loop by e^(−sL). On s = jω its magnitude is 1 and its phase is −ωL radians, or −57.3°·ωL. Because the magnitude does not change, the gain crossover ωc stays where it was and the phase margin falls by exactly ωcL radians. The loop loses stability when that equals the margin it had: the delay margin is L_max = PM/ωc, with PM in radians.

02 / FIELD NOTE

The motor with its gains fixed

The motor lab's default PID (P = 0.22, I = 0.6, D = 0.008 at 20 ms) crosses over at 3.40 rad/s with a phase margin of 58.1°. Each 100 ms of input dead time removes 3.40 × 0.1 × 57.3 = 19.5° there, and the delay margin is 1.014 rad / 3.40 rad/s = 0.298 s. The table gives the frequency workbench's margins and the overshoot of a 20 rad/s step with no load, a step small enough that the voltage never reaches 12 V.

The runs follow the margins: the overshoot roughly doubles with every 100 ms, and at 300 ms, just past the delay margin, the oscillation grows instead of dying out. None of the gains changed.

Dead timePhase marginGain marginOvershoot, 20 rad/s step
None58.1°29.5 dB19.9%
100 ms38.6°10.6 dB37.8%
200 ms19.1°4.8 dB78.8%
300 ms−0.4°−0.1 dBGrowing oscillation
03 / FIELD NOTE

Lower gain buys margin, not speed

Retune instead. With I = D = 0, take the largest P that still leaves about 45° of phase margin at each dead time. The overshoot of the 20 rad/s step then stays near 20%, but the crossover, and with it the speed of the loop, falls steadily. (P alone leaves a steady-state error that grows as P falls; the point here is the crossover.)

The product ωc·L creeps up instead of staying fixed because PM = 180° + ∠G(jωc) − 57.3°·ωcL: as the crossover falls, the motor's own lag at crossover shrinks and leaves more of the phase budget for the delay. Once the delay dominates, the reachable crossover falls in proportion to 1/L. Keep P = 0.655 and set 200 ms instead: the margin is −20.5° and the loop is unstable.

No gain buys this back, because gain moves the magnitude and the delay takes phase. Derivative or lead action adds a bounded amount of phase while raising the high-frequency gain; it shifts the ceiling but does not remove it. A Smith predictor takes the delay out of the characteristic equation when the model is perfect, yet the output still cannot move until L after the input does.

Dead timePPhase marginCrossover (rad/s)ωc·L (rad)Overshoot
50 ms0.65545.1°7.630.3818.6%
100 ms0.4644.9°5.580.5620.4%
200 ms0.30545.3°3.760.7519.7%
300 ms0.2444.7°2.930.8819.4%
500 ms0.17546.0°2.071.0316.4%
04 / FIELD NOTE

Try it

Open the Find the feedback loop experiment, set Target to 20 rad/s and Load torque to 0, and open the frequency workbench. Under Plant additions, step Input dead time through None, 100, 200 and 300 ms and read the phase margin and the overshoot. Then set Integral and Derivative to 0 and pair each dead time with the P from the second table: the phase margin stays near 45° while the gain crossover falls.

05 / FIELD NOTE

Limits

The lab delays the drive voltage in whole 5 ms steps, and the workbench includes that delay exactly in the 20 ms sampled loop, together with the half-sample lag of the hold; it leaves out saturation and load. The 45° target is a choice, not a law, and this motor's own lag is small, so a plant with slower dynamics enters the 1/L regime later. A delay in the measurement path acts on the loop in the same way as an input delay.

TAKEAWAY

Dead time takes ωc·L radians of phase at crossover. Keeping the margin means lowering the crossover, and once the delay dominates, the fastest loop you can build scales like 1/L.

CHECK YOUR UNDERSTANDING: A loop has a phase margin of 50° at a crossover of 2 rad/s. How much dead time can it take before it loses stability?

50° is 0.873 rad, so L_max = 0.873/2 = 0.436 s, provided that crossover is the only one and nothing else changes.

PUT IT TO THE TEST

From reading to a repeatable run.