MiDEN Technical Insights

Common Mode Choke for Server Power Supply

Introduction

Common mode choke for server power supply design sits between two constraints that pull in opposite directions. The emissions limit is hardest to meet in the low megahertz range, where switching harmonics and common mode current crowd together, and a larger Y-capacitor is the cheapest way to buy margin there. The leakage current through those same capacitors is what the residual current device protecting the rack actually sees, and every supply in the rack adds to it. This guide covers where the Class B limit bites, how the Y-capacitor trades against leakage current, and the data to send with an RFQ.

Technical Analysis: Where the Class B Limit Actually Bites

Conducted emission limits for information technology equipment come from CISPR 32, published in Europe as EN 55032, which has largely replaced CISPR 22. Class B applies to equipment intended for a domestic environment. Class A applies to equipment used in a commercial or industrial environment with a defined installation restriction. A server supply sold into a data centre is often taken down the Class A route, but Class B is increasingly requested because it removes the installation restriction and simplifies the documentation.

The conducted limit covers 150 kHz to 30 MHz and it is not a single number. The quasi-peak and average limits are flat through the middle of the range and step back up at the top, and the shape of that curve decides where a design runs out of margin.

The practical consequence is that 500 kHz to 5 MHz is where a server supply usually runs out of margin, and it is not the region where a choke behaves like a pure inductor. A common mode choke is only inductive below its self-resonance; above it, interwinding capacitance dominates and the impedance falls away. Insertion loss measured across the band, at the intended source and load impedance, is a more useful selection figure than a single inductance value.

Impedance balance matters as much as impedance. If the two windings are not closely matched, part of the differential mode load current converts into common mode current, which the choke cannot attenuate because it sees that component as differential. It is one reason a part specified only by inductance can measure well on one board and fail on another.

Design Challenges: Y-Capacitor Versus Leakage Current

Three trade-offs decide whether the filter passes both the emissions scan and the safety review.

• The Y-capacitor is the cheapest attenuation and the largest leakage contributor. Line to earth current through a Y-capacitor rises in direct proportion to capacitance and to line frequency, so the value that buys margin at 500 kHz is the value that spends the leakage budget at 50 Hz. There is no way to take one without the other.
• The leakage budget is shared, not per unit. A rack holding eight supplies leaks roughly eight times as much as one. A 30 mA residual current device protecting a rack branch can be tripped by a set of supplies that are each individually compliant, because the sum crosses the threshold first. Set the per-supply allocation from the number of supplies and from the device protecting the branch, not from the standard alone.
• Insertion loss is measured at small signal far too often. Load current biases the core and reduces the choke's effectiveness in exactly the band where the margin is thin. Ask for the insertion loss curve at the working current and temperature. A curve taken at low current will not predict the rack.

Two further points belong in the drawing rather than in a later conversation. Winding balance and winding-to-winding capacitance both affect how much differential current turns into common mode current, and both are manufacturing parameters rather than datasheet ones. The working voltage between the windings and to the core sets creepage and the insulation system, which is a safety requirement on the AC input stage and has to be stated with the enquiry.

Application Scenarios: 1U to 5.5 kW Rack Supplies

The same selection sequence lands in different places depending on the power level and the input distribution.

• 1U, 3 kW, AC input. The height budget drives the construction more than the electrical requirement does. A low profile or surface mount choke keeps the filter inside the slot, and the Y-capacitor value is usually set by the leakage allocation first and by the emissions margin second.
• 5.5 kW and above. Load current rather than noise sizes the winding. The copper cross section grows, interwinding capacitance grows with it, and the impedance at the top of the band falls. Two stages are common, the second smaller and closer to the bridge.
• Higher voltage DC distribution. Feeding the rack from a higher voltage DC bus removes the line frequency leakage path that a Y-capacitor relies on, so the trade-off changes character and differential mode noise becomes the harder problem. The choke still carries the full load current.

Product Connection: Which MiDEN Family Fits

The AC input filter position is served by three families.

• The common mode choke series covers the input stage, including closed magnetic path constructions that limit the stray field next to the bridge.
• The power inductor series covers the differential mode stage behind the choke, which carries the full load current.
• Custom magnetic components are the route when the leakage allocation, the insulation system or the height budget rules out a standard body.

What these parts are being asked to deliver is set out in the article on common mode choke working principle, and the split between the two filter stages is covered in common mode choke versus differential mode choke. Both are worth reading before the enquiry is written, because the RFQ data only makes sense once the noise path has been separated. The wider component set is listed under magnetic component applications.

Frequently Asked Questions

How much leakage current is allowed in a server power supply?
The limit depends on the equipment class, the earthing arrangement and the safety standard the product is certified against, so it has to be taken from the applicable standard for that product. In a rack the practical ceiling is usually lower than the standard allows, because the leakage of every supply adds up and the residual current device protecting the branch is what actually trips.

What does Class B require from a common mode choke?
Enough insertion loss across 500 kHz to 5 MHz, measured at the working current and temperature. That band is where the quasi-peak limit is at its tightest and where the choke stops behaving like a pure inductor, so insertion loss in the band is a more useful figure than nominal inductance.

Can a larger Y-capacitor fix the emissions margin?
It buys attenuation at the low end and spends leakage current at the same time, and the leakage it adds is shared with every other supply in the rack. It is a compliance decision and a leakage decision at once, so it normally has to be made together with the choke rather than after it.