Latency in Closed-Loop Motor Control: Why Microseconds Matter

1. The closed-loop chain

A motor under closed-loop control is a continuous round-trip: the drive commands a current, the motor moves, the encoder measures the new position, the feedback travels back to the drive, and the drive computes the next command. The total latency around this loop sets a hard upper bound on how fast the controller can respond.

Each link in the chain contributes:

• Encoder edge to capture register — typically nanoseconds for an HTL/TTL channel
• Capture to feedback transmission — wired: nanoseconds; wireless: microseconds to milliseconds
• Feedback delivery to drive ISR — depends on drive architecture, typically tens of microseconds
• Drive control loop computation — fixed by drive hardware, typically 50–250 µs
• PWM update to output stage — typically one PWM period

Most of these are determined by the drive and are not negotiable. The single component the system designer can choose is the feedback delivery path. That makes it the right place to optimize.

2. Latency budget breakdown

For a high-performance velocity loop running at 8 kHz update rate (125 µs sample period), the typical industrial budget looks roughly as follows:

The wired column sets the benchmark. A high-quality wireless link adds roughly 0.5–1 ms — enough to be felt in a fast servo, but tolerable for industrial motion control at the typical 1–10 kHz update rates. A poorly engineered wireless link adds milliseconds with substantial worst-case spikes, which forces the controller to either tolerate large phase loss or detune until it can.

3. Effect on PI controller stability

A velocity PI controller has two tuning parameters: proportional gain (K_p) and integral gain (K_i). The achievable values are bounded by the requirement that the closed-loop system remain stable in the presence of the round-trip delay τ.

For a first-order plant with delay, the classical Bode-stability rule of thumb gives a maximum useful proportional gain that scales as 1/τ: doubling the loop delay cuts the achievable gain in half. That gain reduction directly translates to:

• Slower response to load disturbance — the motor takes longer to recover speed when load changes
• Lower disturbance rejection — short transients are not flattened
• More overshoot on reference change — step-response settling time grows
• Reduced robustness — the same plant tolerates less variation in load inertia or friction

A concrete example

Consider a velocity loop with 1 ms feedback latency tuned to give critical damping with K_p = 0.5. With 5 ms latency on the same plant, K_p must drop to roughly 0.1 to maintain the same damping. Step response settling time grows from ~5 ms to ~25 ms. Disturbance rejection (the response to a sudden load change) is ~5× slower. The system still works — it is just a fundamentally lower-bandwidth controller.

4. Why the worst case matters more than the average

Latency in a wireless link is rarely a single number. It is a distribution: a typical case, a 99th percentile, a 99.99th percentile, and (in poorly designed systems) outliers in the tens of milliseconds caused by retries or channel switches.

A drive engineer cannot tune for the average. The controller must remain stable at the worst case observed in the system, because instability at the 99.99th percentile means a fault stop or an oscillation event every few hours. So the question to ask of a wireless encoder vendor is not "what is the typical latency" — it is "what is the 99.99th percentile latency over 24 hours under normal industrial RF conditions", and "is that figure bounded or unbounded?"

An unbounded latency distribution — one with rare but extreme tails — is unsuitable for closed-loop control regardless of the average. A bounded distribution with a hard ceiling (say, "always under 1 ms") permits the engineer to tune for that ceiling and forget about it.

5. Practical impact on motor performance

The downstream effects of a slow or jittery feedback loop show up in ways that are often blamed on the motor or the drive:

• Speed ripple at constant setpoint — the controller cannot respond fast enough to small load variations
• Audible hum or vibration — sub-resonant oscillations from limit-cycle behavior near the stability boundary
• Visible overshoot on acceleration — step responses that "ring" before settling
• Mechanical wear from repeated correction — the controller chasing its own tail
• Tuning drift — gains that worked at commissioning need re-tuning months later as the plant ages

Each of these is a symptom of insufficient control loop bandwidth. The instinct is to upgrade the drive or the motor; the actual fix is often to give the existing controller better feedback.

6. What "low latency" should mean quantitatively

The industry uses the term loosely. We recommend the following operational definition for industrial wireless encoder feedback:

7. How wireless adds latency — and how a good design minimizes it

Three architectural choices dominate wireless encoder latency:

1. Protocol overhead. A generic Wi-Fi stack (CSMA/CA, retries, ACKs) adds hundreds of microseconds in the best case and tens of milliseconds in the worst. A purpose-built TDMA frame for periodic encoder data can run at fixed sub-millisecond cadence with no retries needed under normal RF conditions.
2. Buffering depth. Anything longer than a single sample period of buffering on the TX or RX side is wasted latency. WENC2 pipelines encoder edges with no application-layer buffering — every edge moves as soon as it arrives.
3. Retry policy. Aggressive retries trade latency for packet success. For closed-loop control, a single missed sample is preferable to a 5 ms-late sample. The protocol must know which to optimize for.

8. WENC2's measured latency

WENC2 ships with a dedicated TDMA protocol on a CE-certified 5 GHz dual-band module. Measured end-to-end latency (TX encoder edge → RX output edge) under normal industrial RF conditions:

• Median: 0.4 ms
• 99th percentile: 0.7 ms
• 99.99th percentile: < 1 ms
• Hard ceiling (single-sample loss policy): 1.0 ms

The hard ceiling is enforced by design: a sample that does not arrive within the 1 ms window is dropped, not retried. This is the correct behavior for closed-loop control, and it is the property that lets the drive engineer tune as if the feedback were wired.

9. What to measure when evaluating a wireless encoder

If you are testing a candidate wireless encoder link on your bench, here is the bare-minimum measurement set:

1. Loop-back latency — feed a signal into the TX encoder input and measure the delay to the RX output with an oscilloscope. Capture at least 1000 events.
2. Latency histogram — bin those 1000 events. The shape of the distribution tells you everything. A tight, narrow peak is good. A long tail is a warning.
3. Jitter under interference — repeat the measurement with a smartphone in Wi-Fi tethering mode placed 1 m from the unit. The link should degrade gracefully, not catastrophically.
4. 24-hour run — log packet success and worst-case latency over a full 24 hours. The 99.99th percentile is the number that matters.

10. Conclusion

Closed-loop motor control is a system-level discipline. The drive, the motor, the encoder and the wiring all contribute to the achievable performance. Replacing wired feedback with a wireless link changes only one component of that system, but it changes it in a way that interacts with everything else.

A wireless encoder link with sub-millisecond bounded latency is a transparent replacement for wired encoder feedback. A link with unbounded latency, or with a worst-case in the multi-millisecond range, is not — it is a different system that requires a detuned controller and accepts lower performance.

The right specification is short and unforgiving: under one millisecond, bounded, no retries past the deadline. That is the bar WENC2 was designed to meet, and the bar against which any candidate wireless encoder should be measured.