Why 5 ms Matters: Dynamic Response Time of Active Harmonic Filters

Summary. Active harmonic filter response time is the single parameter that decides whether compensation works on a real load or only on a laboratory bench. A filter that detects and cancels distortion within 5 ms can follow a welding inverter, an EV fast charger ramp or a drive acceleration, while a 40 ms device only corrects the average of a transient that has already passed. This article explains what response time actually measures, why the difference between 5 ms, 20 ms and 40 ms changes the THDi you will see in the field, how DSP control architecture and CT bandwidth set the real limit, and how to commission and verify dynamic response with a load-step test. It also covers which applications genuinely need 5 ms performance, and where a slower unit is a perfectly sound engineering choice.

What Does "Response Time" Mean on an Active Harmonic Filter Datasheet?

Response time is the interval between a change in the load current and the moment the filter output current has fully matched the new harmonic demand. It is a closed-loop figure, not a switching figure. Three delays stack up inside it: measurement delay (current transformers, signal conditioning and the ADC sampling window), computation delay (FFT or synchronous-reference-frame extraction plus the control law), and actuation delay (PWM update, IGBT switching and output filter dynamics). A vendor claiming "5 ms response" should mean the complete chain, measured from a step change at the CT primary to 90% compensation at the filter terminals.

Typical figures across the market fall into four bands. Fixed passive filters have no response time in this sense at all, because their impedance is frequency-defined, not load-defined. Older analogue or low-speed digital AHFs sit at 20–40 ms, usually because they compute harmonic content over a full-cycle window. Mainstream digital units achieve 10–20 ms. High-performance units built around a fast DSP with a sub-cycle detection window reach 5 ms or better, with full compensation typically completed inside 20 ms and a settling time under one fundamental period.

The measurement window matters more than most buyers realise. A 20 ms window spans exactly one cycle at 50 Hz, so the controller cannot report a change until the cycle has finished — the response is inherently at least one cycle late. Reducing the effective window requires either a shorter transform length with interpolation, or a time-domain control structure such as deadbeat or repetitive control running at a 10–20 kHz sampling rate. That is why the two specifications most correlated with genuine 5 ms performance are DSP clock speed and switching frequency, not the size of the enclosure.

A second definition problem is overshoot versus settling. A controller can reach the new current in 4 ms and then oscillate for 60 ms, and the datasheet will still say "4 ms". Ask separately for rise time, settling time and maximum overshoot, and ask what happens on a 100% step from zero. Units with poorly damped resonant control show dramatic overshoot on zero-to-rated steps, which in the field appears as current spikes that trip upstream protection. The best-performing designs combine a fast detection window with a feed-forward term from the load current, so the loop anticipates the step rather than only reacting to it after the error accumulates.

Why 5 ms Performance Changes the Outcome on Real Loads

1. Transient loads never settle. Welding lines, induction heaters, press shops and spot welders change their harmonic signature several times per second. A 40 ms filter chases a spectrum that has already moved on, so the measured THDi stays stubbornly high even though the unit is "working". On a 200 kVA welding bay, replacing a 40 ms unit with a 5 ms unit commonly halves residual THDi — not because the algorithm is smarter, but because it is fast enough to be relevant.

2. EV charging ramps are steep. A DC fast charger can step from 20% to full load in well under 100 ms, and a charging hub with six stalls produces overlapping steps. Slow compensation appears as flicker and as transformer audible noise. A charging-station AHF with sub-cycle response keeps the point-of-connection current clean through the whole ramp, which matters increasingly because utilities now meter harmonic emission at the charger feeder.

3. Renewable inverters interact with the filter. Grid-tied PV and storage inverters run their own fast control loops. If the AHF response is slower than the inverter's own dynamics, the two controllers can exchange energy at low frequency, producing oscillation and, in the worst case, nuisance tripping at the point of interconnection. A fast-response AHF for photovoltaic systems is tuned to sit outside the inverter bandwidth, so the two devices cooperate instead of fighting.

4. Sensitive loads have no tolerance for residual distortion. Data halls, semiconductor lines and medical imaging systems are protected by UPS units that see harmonic current as extra losses. Slow compensation leaves short but repeated distortion bursts that accelerate capacitor ageing. Selecting a 100A data center AHF with 5 ms response and per-order logging turns an unquantified risk into a measurable, reportable number.

5. Fast response enables reactive and balancing functions to share one device. Power factor correction and three-phase balancing are step events too. If the reactive loop is slow, voltage support lags the load and the utility sees flicker; if it is fast, the same hardware satisfies PF, imbalance and THD requirements simultaneously. That is why response time is now quoted alongside PF range in tender documents, not treated as a footnote.

6. Compliance windows are getting shorter. Harmonic limits are increasingly assessed over short observation intervals rather than ten-minute averages, because distributed resources and fast loads make averages misleading. A device that only performs on average will fail a short-window assessment even when its long-term numbers look acceptable. Speed therefore has a compliance dimension: the emission measured in any 200 ms window must stay inside the limit, which is only achievable when compensation keeps up with the load.

How to Verify and Specify Dynamic Response in Practice

Step 1 — Ask for the definition, not just the number. Request the test method: step size (usually 20–100% of rated current), measurement point, and whether the figure is 90% settling or full settling. A reputable supplier will also state the sampling rate and switching frequency, because those two numbers set the physical floor.

Step 2 — Run a load-step test at commissioning. With a Class A analyser and a current probe on the filter output, switch a known load bank or a drive from idle to rated current and capture the waveform. Expect the filter current to reach its new steady value within 5 ms and to be stable within 20 ms, with no overshoot beyond 10% and no sustained oscillation. Record the trace in the handover pack; it is the only objective evidence a consultant will accept.

Step 3 — Match CT bandwidth to the controller. A 5 ms controller fed by a 100 Hz-bandwidth CT cannot see the 25th order. Specify CTs with sufficient bandwidth and correct burden, place them on the load side of the filter so the controller regulates grid current rather than its own output, and verify phase polarity at commissioning. Mismatched polarity is the most common cause of a fast filter appearing to respond slowly.

Step 4 — Tune gain for stability, not for speed. Every fast loop trades speed against stability. On a 30A module serving a mixed commercial board, a moderate gain with 5 ms response is preferable to maximum gain with oscillation. On heavy industrial networks such as a 690V 200A mining system, the priority shifts to robustness: wide input tolerance, IP54 protection and stable operation under voltage dips matter more than shaving a millisecond.

Step 5 — Size for headroom so speed is available when needed. A filter running at 95% of rated current has no reserve to respond to a step. Keeping 20–30% headroom, or using a paralleled modular frame such as a 75A floor-standing cabinet, means the dynamic capability is genuinely available during peak load rather than only on paper.

Step 6 — Coordinate multiple modules and controllers. In a multi-module system, controller communication adds latency that no single-unit datasheet mentions. If four modules share one CT, the master must distribute the current reference fast enough that the modules act together; a slow bus turns a 5 ms system into a 15 ms one. Specify the update rate of the module-to-module link, confirm the paralleling method (analogue reference, CAN, RS485 or Ethernet), and test the full system with a step load after all modules are commissioned — not just one module on the bench.

Step 7 — Confirm speed survives derating. Fast response is only useful while the module remains within its thermal envelope. Ask for the derating curve above 40 °C ambient, check that cabinet airflow is not obstructed after installation, and note that a unit limited by temperature will quietly reduce output and therefore lose its dynamic benefit during the hottest production hours. Where ambient conditions are severe, choose a cabinet-mounted frame with forced cooling rather than a sealed wall unit.

One practical caution: response time is often marketed in isolation, but buyers should assess it together with harmonic order coverage and per-order selectivity. A filter that responds in 5 ms across the 2nd to 50th order range is useful; one that responds in 5 ms on a narrow band of orders is not. Request both specifications on the same datasheet and confirm they are measured under the same load condition.

FAQ

Is a 5 ms response time really necessary?

Only where the load changes faster than about 50 ms. Welding, EV fast charging, PV inverters and press lines need it. For pumps, fans, chillers and steady conveyor drives, a 10–20 ms unit delivers essentially the same final THDi at lower cost.

How is response time measured on a datasheet?

Most suppliers quote 90% settling after a step change from 20% to 100% of rated compensating current, measured at the filter terminals with a Class A analyser. Ask for the step size, measurement point and whether settling or full response is quoted.

Does a higher switching frequency always mean faster response?

Not automatically. Switching frequency improves current-loop bandwidth and reduces ripple, but the detection window and control law dominate overall response. A 20 kHz unit with a full-cycle FFT can still be slower than a 10 kHz unit using sub-cycle time-domain control.

Can response time be improved after installation?

Sometimes. Firmware updates, gain retuning and correcting CT polarity or placement can recover a full cycle of delay. If the hardware uses a slow processor or a full-cycle detection window, no tuning will reach 5 ms and the unit must be replaced.

Why does my filter show high THDi despite a fast response claim?

Usually because the CT is downstream of the filter, the unit is running without headroom, or the measurement point is downstream of uncompensated local loads. Re-check CT placement, current headroom and the compliance point before blaming the controller.

How does response time interact with reactive power compensation?

They share the same loop. Fast response gives tighter voltage support and better flicker suppression, but aggressive reactive gain can cause hunting on weak grids. Commissioning should set harmonic and reactive gains separately and verify both under step load.

Conclusion

Response time separates an active harmonic filter that performs in a plant from one that only performs in a brochure. The 5 ms class exists because modern loads — welding inverters, DC fast chargers, PV and storage converters — change faster than a full-cycle controller can follow. Verify the definition, test it with a load step at commissioning, keep 20–30% current headroom, and pair speed with broadband 2nd–50th order coverage so the capability actually applies.

Need a dynamic response guarantee, not just a datasheet claim? Send us your load profile and step-test requirements, and we will propose a matching frame size with measured response traces and a commissioning checklist. OEM and ODM projects with custom control parameters are welcome. Contact our engineering team for a quotation or technical proposal.