Блог о качестве/конденсаторах

Низкоиндуктивное шунтирующее измерение импульсов УФ-лампы частотой 200 кГц

Краткое содержание · Основные инженерные решения

A 10 nH sense loop adds 10 V of error to a 50 mV signal at 1000 A/µs, and at 500 kHz that loop looks like 31 mΩ against a 1 mΩ shunt. Kelvin shunts, a 1 cm² sense triangle and Class I filter capacitors with ESL below 5 nH keep the anti-alias corner fixed.

2026-10-28·21 min read· Читать 0 раз

Низкоиндуктивное шунтирующее измерение импульсов УФ-лампы частотой 200 кГц

Высоковольтные компоненты HVC в каскаде трансформатора тока
Высоковольтные компоненты HVC в каскаде трансформатора тока

Инженерный ответ

Hold sense-loop inductance below 5 nH. At 1000 A/µs a 10 nH loop adds 10 V of error to a 50 mV signal. At 500 kHz a 10 nH lead presents 31 mΩ, which is 31 times a 1 mΩ shunt. Use four-terminal Kelvin shunts, a one-square-centimetre sense triangle, and Class I filter capacitors with tanδ below 0.1%.

  • Sense from dedicated Kelvin taps, never from the load terminals
  • Keep the shunt, filter capacitor and amplifier input in one small triangle
  • Measure loop inductance on the first prototype and target below 5 nH
  • Set the anti-alias corner from the RC product, not from amplifier bandwidth
  • Verify gain and phase with a 50 A, 100 ns calibration pulse

Резюме

A UV curing supply measures 100 A pulses with nanosecond edges, and every nanohenry in the sense path turns into voltage error. This article works through the inductive error term, the reactance of a 10 nH lead at 500 kHz, the Kelvin connection and the RC filter corner. It then gives six design rules that hold a 200 kHz sense loop within a few percent.

1. Введение

A UV curing power supply needs real-time current in three places: the inverter switches, the trigger transformer and the lamp arc. The inverter current runs at 100–500 kHz with nanosecond edges.

The ignition pulse reaches tens of amperes with a sub-microsecond risetime, and the arc current collapses at breakdown. One sensor covers all three nodes, and that sensor is the shunt resistor.

The shared problem is inductance. A sense loop converts a valid current signal into a voltage error, and the error grows with the slew rate of the pulse it measures.

This article covers the error arithmetic, a worked example at 200 kHz, four field pain points, and the layout and component rules that keep the error inside a few percent.

2. Основной механизм

2.1 The inductive error term

The sense path behaves as a series network of shunt resistance and loop inductance. Its output voltage is:

ΔV = L_loop × di/dt

where ΔV is the inductive error voltage, L_loop the total inductance of the current path plus the sense loop, and di/dt the current slew rate of the pulse.

At 1000 A/µs a 10 nH loop produces 10 V. The wanted signal is 50–100 mV, so the error is 100–200 times larger than the measurement itself.

Loop geometry sets L_loop. A wide return path around the shunt behaves as a single-turn air-core inductor, and it picks up the very pulse it is measuring.

2.2 Reactance of the lead at 500 kHz

Lead inductance also distorts the reading at the switching frequency, even when the pulse edge is not the issue:

X_lead = 2πf × L_lead

where X_lead is the reactive impedance, f the frequency component being measured, and L_lead the series inductance of the shunt lead and the wiring.

A 10 nH lead at 500 kHz gives 31 mΩ. Against a 1 mΩ shunt that reactance is 31 times the resistance, so the reading carries a 90° phase error.

A phase error of that size eats the phase margin of the current loop, and the loop starts to ring at the switching frequency.

2.3 The Kelvin connection

Four-terminal connection separates the current path from the voltage sense path. The sense wires then carry almost no load current:

V_sense = I × R_shunt + L_body × di/dt

where V_sense is the voltage at the Kelvin taps, I the load current, R_shunt the shunt resistance, L_body the self-inductance of the resistor body, and di/dt the current slew rate.

Only the resistor body remains in the sense branch, so the lead and loop terms leave the equation. A planar foil or thick-film body with short, wide terminations holds L_body in the low nanohenry range.

On real hardware, moving from two-terminal to four-terminal sensing cuts the total sense inductance by roughly an order of magnitude.

2.4 Filter corner and self-resonance

The anti-alias filter sits between the Kelvin taps and the amplifier input. Its corner comes from the filter resistor and capacitor:

f_c = 1 / (2π × R_filter × C_filter)

where f_c is the corner frequency, R_filter the series filter resistance, and C_filter the filter capacitance at the amplifier input.

The capacitor is not ideal, and its own inductance creates a second frequency of interest:

f_SR = 1 / (2π √(L_ESL × C_filter))

where f_SR is the self-resonant frequency, L_ESL the equivalent series inductance of the capacitor, and C_filter the filter capacitance.

A 10 nF capacitor with 10 nH of ESL resonates at 16 MHz. Above that frequency the capacitor behaves as an inductor, and the filter amplifies the noise it was meant to remove.

Ignition harmonics land in the 5–20 MHz band, so an ESL-heavy filter turns them into phantom overcurrents that trip the protection circuit.

Sense Error vs Loop Inductance 0 5 10 15 20 Error voltage (V) 5 10 15 20 Sense loop inductance (nH) 1000 А/мкс 500 А/мкс 15 nH at 500 A/µs gives 7.5 V on a 50 mV signal Alarm level of the protection comparator

Fig. 1 — A 15 nH sense loop turns a 500 A/µs trigger edge into 7.5 V of error, and a 5 nH loop holds the same edge near 2.5 V.

3. Проблемы проектирования

Pain Point 1: Inductance-dominated readings

The trigger pulse is 50 A with a 100 ns risetime, so di/dt reaches 500 A/µs. Through a 15 nH sense path that gives 7.5 V of inductive error on top of a 50 mV signal.

The ADC saturates, the protection circuit false-trips, and the lamp does not start. The schematic looks correct because inductance does not appear on it.

Pain Point 2: CMRR loss from asymmetric wiring

The two sense wires run at different lengths and reference different ground points. The 400 V common-mode switching step then becomes a differential error of hundreds of millivolts at the amplifier input.

The amplifier output modulates at the switching frequency, the current loop follows it, and the cured product shows banding across the web.

Pain Point 3: Filter capacitor resonance

A 10 nF filter capacitor with 10 nH ESL has its self-resonance at 16 MHz. At 500 kHz the part is comfortably capacitive, so bench tests pass.

At the 5–20 MHz ignition harmonics the same part resonates. Noise is amplified, and the protection circuit sees overcurrents that no real load produced.

Pain Point 4: Thermal and aging drift of the sensing chain

The shunt heats under pulsed current, and its resistance drifts with the I²R duty cycle. The filter capacitor ages at the same time, so the corner frequency moves.

Factory calibration then no longer matches the hardware. Thresholds set with tight margins begin to false-trip, or they miss real overcurrents.

4. Этапы проектирования

Strategy 1: Use four-terminal Kelvin shunts

Route the load current through the main terminals and sense from dedicated Kelvin taps. That removes the load current from the sense loop and cuts the inductive error by about a factor of ten.

Choose a planar low-profile толстопленочный резистор with short, wide terminations, and hold its body inductance below 5 nH. Where a wirewound shunt cannot reach that figure, a flat-element резистор выборки это альтернатива.

Strategy 2: Minimize the sense loop area

Place the shunt, the filter capacitor and the amplifier input inside a small triangle on a dedicated ground island. Every square centimetre of enclosed area adds roughly 10 nH.

Measure the finished loop with a network analyzer on the first prototype. A reading below 5 nH holds the error near 2.5 V at 500 A/µs, against 7.5 V at 15 nH.

Strategy 3: Integrate the RC filter with Class I capacitors

Используйте N4700 Class I dielectric for the anti-alias filter. Its tanδ stays below 0.1%, so the filter neither heats nor shifts the measurement phase.

A temperature coefficient of ±30 ppm/°C holds the corner frequency fixed across the cabinet temperature swing, so the phase lag of the measurement does not drift through the day.

Strategy 4: Specify the amplifier for the pulse shape

Choose an amplifier with bandwidth and slew rate at least five times the highest frequency component of the measured pulse. Its CMRR must be 80 dB or better at the switching frequency.

Verify the rejection on the bench with the real switching waveform. Datasheet CMRR is quoted for a balanced source, and a real sense loop never presents one.

Strategy 5: Calibrate in situ with a known pulse

During production test, inject a calibrated 50 A, 100 ns pulse and record the ADC reading. Store the gain and phase correction in the controller, then repeat the calibration every 5000 operating hours.

That schedule tracks shunt drift and capacitor aging together, which is the pair that moves the reading in Pain Point 4.

Strategy 6: Simulate the sense path before layout

Model R_shunt, L_loop, C_filter and the amplifier input impedance in SPICE with the real di/dt. The simulation predicts the overshoot and ringing that would otherwise appear only on the bench.

Fixing ringing in simulation costs engineering hours. Fixing it after the board is potted costs a respin and a week of schedule.

Параметр компонентаТипичное значениеУсловие теста
Номинальная мощность (серия HVR)50 Вт при 70 °Cалюминиевый корпус, свободный поток воздуха
Значение сопротивления10 кОм ±5%эталонная температура 25 °C
Рабочее напряжение10 кВ переменного/постоянного тока(CIJ)
Устойчивость к импульсам30 кВ при 10 мкс10 импульсов, класс IEC 60110
Температурный коэффициент (резистивный)±50 частей на миллион/°Cот −55 °C до +155 °C
сопротивление изоляциисопротивление выше 1 ГОм при напряжении 1000 В постоянного токаминимальное смещение 25 мм
Возможность перенапряжения100 А при 1 мснеиндуктивная проволочная обмотка

5. HVC N4700 capacitors for the sense filter

Three requirements drive the choice at this node: ESL low enough to keep the filter capacitive across the harmonic band, loss low enough that the filter does not heat, and capacitance stable enough that the corner stays put. HVC builds Class I parts for that set, from a 6,000 m² owned factory in Dongguan, across 1 kV to 100 kV.

Шесть параметров, которые здесь имеют значение.

  • Low-ESL Class I capacitors in compact disc and 441 screw-type packages keep the filter self-resonance above 30 MHz, so the sense filter stays capacitive over the whole switching and harmonic band.
  • tanδ below 0.1% at 1 kHz means the filter adds no heating and no phase distortion into the measured pulse.
  • A ±30 ppm/°C temperature coefficient holds the anti-alias corner fixed, so the measurement phase lag does not drift with cabinet temperature.
  • A 1–100 kV voltage range covers the low-voltage sense filter and the high-voltage trigger and ballast stages, so one qualified family serves the whole supply.
  • Guaranteed PD below 5 pC at 1.2× rated voltage keeps the high-voltage sites that share the sense board from eroding and disturbing the measurement reference.
  • HVC's 40 kV / 441 screw-type capacitor replaced a first-tier international brand in field-service programmes where current and voltage readings feed fine controls. дверная ручка конденсаторы и диски соответствуют параметрам Vishay и высоковольтные серии TDK.

Доставка и поддержка

Standard parts ship in 4–8 weeks, and matched filter sets in 6–10 weeks. HVC applications engineering reviews the sense path, the loop inductance target and the RC corner against your waveform.

Send the pulse waveform, shunt specifications, amplifier choice and layout constraints. HVC returns a low-inductance sense design review, Class I filter recommendations and SPICE-ready models within five working days, with samples in two weeks.

6. Часто задаваемые вопросы

Q1: How much error does 10 nH of stray inductance add?

It depends on di/dt. From ΔV = L_loop × di/dt, a 10 nH loop adds 10 V at 1000 A/µs and 5 V at 500 A/µs, against a wanted signal of 50–100 mV. Loop area, and not shunt tolerance, sets the accuracy of a fast pulse measurement.

Q2: Why does a 1 mΩ shunt read like an inductor at 500 kHz?

Because X_lead = 2πfL_lead gives 31 mΩ for 10 nH at 500 kHz. The reactance is 31 times the resistance, so the lead dominates the measured voltage and the reading carries a 90° phase error that corrupts loop stability.

Q3: What does a Kelvin connection actually remove?

It removes the load current from the sense branch. The sense wires then carry only the small amplifier input current, so the lead and loop inductance terms leave the equation. What remains is the body resistance plus the body self-inductance, and a planar low-profile resistor holds that second term small.

Q4: How do I choose the anti-alias corner?

Set f_c = 1/(2πR_filter×C_filter) above the highest real component of the pulse, typically five to ten times the fundamental, and below the first sampling alias. A 50 Ω filter resistor with 1 nF puts the corner at 3.2 MHz, which clears both the 500 kHz switching term and the ignition edge.

Q5: Why does filter capacitor ESL matter more than its capacitance?

Because the ESL sets where the filter stops working. A 10 nF part with 10 nH resonates at 16 MHz, and above that point it amplifies noise instead of shunting it. Ignition harmonics reach 20 MHz, so a low-ESL part with self-resonance above 30 MHz keeps the whole band covered.

Q6: How often should the sense chain be recalibrated?

Every 5000 operating hours, using the same 50 A, 100 ns pulse as the production test. That interval catches shunt resistance drift from I²R heating and the capacitor aging that moves the corner, and it is short enough that protection thresholds keep their margin.

7. Рекомендации

  1. Э. Куффель, В.С. Заенгль, Дж. Куффель, Основы высоковольтной техники, 2-е изд., Ньюнес, 2000.
  2. Мохан, Н., Унделанд, Т.М., и Роббинс, В.П. Силовая электроника: преобразователи, применение и дизайн, 3-е изд. Хобокен, Нью-Джерси: John Wiley & Sons, 2003.
  3. Уведомление 2 MIL-HDBK-217F. Прогнозирование надежности электронного оборудования, 1995.
  4. МЭК 60384-1:2021, Фиксированные конденсаторы для использования в электронном оборудовании — Часть 1: Общие технические характеристики, 6-е изд. Женева: МЭК, 2021.
  5. МЭК 61071:2017, Конденсаторы для силовой электроники, Изд. 2.0. Женева: Международная электротехническая комиссия, 2017.
  6. ГБ/Т 7354-2018, 高电压试验技术 局部放电测量. Источник: 中国标准出版社, 2018.
  7. А. Марискотти и Л. Ваккаро, Катушка Роговского для высоковольтных применений, в сборнике трудов конференции IEEE Instrumentation and Measurement Technology Conference 2008 года, стр. 1069-1073, 2008. https://doi.org/10.1109/imtc.2008.4547197.
  8. Х. Накао, Ю. Ёнезава, Т. Сугавара, Ю. Накашима и Ф. Курокава, Метод онлайн-оценки деградации электролитических конденсаторов для прогнозирования отказов импульсных источников питания с цифровым управлениемIEEE Transactions on Power Electronics, том 33, № 3, стр. 2552-2558, 2018. https://doi.org/10.1109/tpel.2017.2691048.
  9. США 6,937,027, Гибридный датчик для обнаружения высокочастотных частичных разрядов, Hanyang Hak Won Co., Ltd., выдано в 2005 г., https://www.freepatentsonline.com/6937027.html.
  10. Использование низких напряжений для проверки высоковольтных конденсаторов (ток утечки).Форум сообщества EEVblog по электронике, https://www.eevblog.com/forum/testgear/using-low-voltage-levels-to-test-hv-capacitors-leakage-current/ .
  11. Команда инженеров по разработке приложений HVC, Керамический конденсатор высокого напряжения Тeхничeскиe рeсурсыhttps://www.hv-caps.com
Для этого приложения необходим проверенный компонент?

Пришлите напряжение, емкость, частоту, монтажный чертеж и количество. Компания HVC в течение 48 часов порекомендует компонент класса N4700 с данными проверки на частичные дыры и отчетами о сроке службы до 1000 часов.

Отправить спецификацию материалов / Запросить техническое предложение →