Conducted EMI Troubleshooting in Switch-Mode Power Supplies: Analysis, Modeling and Countermeasures

Why Conducted EMI Troubleshooting in SMPS Demands Physics-Based Analysis

Conducted electromagnetic interference (EMI) remains one of the most persistent barriers to regulatory compliance and field reliability in switch-mode power supplies (SMPS). Unlike thermal or efficiency issues, EMI failures rarely manifest as catastrophic faults — instead, they appear as subtle, frequency-specific violations during pre-compliance testing, often only after thermal stabilization or under specific grounding conditions. This unpredictability leads many engineers to default to “brute-force” troubleshooting: cycling through dozens of filter components, layout tweaks, and grounding changes until the emissions curve dips below CISPR 32 or FCC Part 15 limits.

But experience across thousands of SMPS designs confirms a stark reality: the theory-guided approach — rooted in noise source identification, parasitic path modeling, and Ohm’s Law–based impedance analysis — reduces average debug time by over 80% and cuts prototype iterations by half. When no EMI receiver is available, high-fidelity circuit simulation (e.g., Saber-based conducted EMI modeling) can validate both the root cause and the proposed countermeasure before any hardware change. This article synthesizes field-proven principles, a real-world 90 W PFC + flyback case study, and actionable design rules — all grounded in the single unifying mechanism that governs >90% of conducted EMI failures: the need for a controlled, low-impedance return path for common-mode current.

The Universal Mechanism: Low-Impedance Return Paths Control CM Noise

At its core, conducted EMI compliance hinges on one physical law applied to parasitic networks: current follows the path of least impedance. In SMPS, the LISN (Line Impedance Stabilization Network) does not “measure noise” — it measures the voltage developed across its 50 Ω impedance when common-mode current flows *through* it. Therefore, reducing conducted emissions means ensuring that CM current avoids the LISN entirely — by giving it a lower-impedance alternative route back to its source.

Three seemingly distinct techniques — heatsink grounding, transformer shielding, and Y-capacitor placement — operate through this identical principle.

Heatsink Grounding: Short-Circuiting the CM Loop

In a flyback converter, the MOSFET drain exhibits extreme dV/dt (often >50 V/ns). A parasitic capacitance C₁ exists between the MOSFET die and the heatsink; another, C₂, exists between the heatsink and protective earth (PE). If the heatsink is floating, CM current i_CM flows:
MOSFET drain → C₁ → heatsink → C₂ → PE → LISN → AC source.
This path injects full CM current into the LISN.

Grounding the heatsink directly to the primary-side quiet ground (not PE!) creates a local shunt:
MOSFET drain → C₁ → heatsink → primary ground → source.
Now i_CM bypasses the LISN entirely. Measured reductions exceed 20 dB at 1–30 MHz — with zero cost, zero efficiency penalty, and no added components.

Transformer Shielding: Reducing the Coupling Source

The transformer’s inter-winding capacitance is the dominant CM coupling path from the high-dV/dt primary side to the secondary. An electrostatic shield (copper tape or foil, grounded to primary ground) inserted between windings breaks this capacitive link. Crucially, the shield must be connected *only* to the primary ground — connecting it to secondary ground reintroduces a CM path.

Field data shows: a well-designed Y1-rated Y-capacitor alone delivers ~20+ dB of CM attenuation at 1–30 MHz. By contrast, optimizing winding geometry (e.g., triple-insulated wire, interleaving) typically yields only 10–15 dB — and often at the cost of increased leakage inductance or reduced efficiency. Shielding is thus the highest-margin first step.

Noise Path Modeling: Applying Ohm’s Law to EMI

Treating EMI as an impedance network transforms troubleshooting from art to engineering. For conducted emissions, two paths dominate:

• Differential-mode (DM) path: Formed by the pulsating triangular input current flowing through the bulk capacitor, PCB traces, and LISN. Dominates below ~1 MHz. Governed by series impedance: higher series inductance (e.g., CM choke) blocks DM current.
• Common-mode (CM) path: Originates at high-dV/dt nodes (MOSFET drain, transformer primary), couples capacitively to secondary, then returns via load-to-earth capacitance, Y capacitors, or output chokes to the LISN. Dominates above ~1 MHz. Governed by parallel current division: lower impedance in the “bypass” branch diverts current away from the LISN.

The goal is simple: minimize V_LISN = I_CM × Z_LISN. Since Z_LISN is fixed at 50 Ω, reduction requires lowering I_CM — achieved by either:

• Reducing the noise source voltage (e.g., slowing gate edges — but increases switching loss),
• Raising the coupling impedance (e.g., reducing transformer C_ps — limited by safety clearance), or
• Lowering the impedance of the preferred return path (e.g., increasing Y-capacitance or removing series ferrites).

Temperature dependence adds nuance: as temperature rises, aluminum electrolytic capacitor ESR falls, lowering the impedance of unintended CM paths (e.g., bulk cap → chassis → LISN). This explains why some units pass cold but fail hot — and why “fixes” that work at room temperature may degrade in operation.

Case Study: Solving a 15–16 MHz Conducted Failure After Warm-Up

A 90 W PFC + flyback supply failed CISPR 32 Class B conducted emissions at 15–16 MHz — but only after 30 minutes of operation and only when the output ground was connected to earth. Initial measurements showed >6 dB margin at room temperature, collapsing to −2 dB at operating temperature.

Root Cause Diagnosis

The grounding dependency was the first clue: CM noise requires a closed loop involving earth. The failure frequency (15–16 MHz) placed it firmly in the CM-dominated band. The thermal sensitivity pointed to parasitic impedance shifts — likely falling ESR in the bulk capacitor or output electrolytics, increasing CM current flow through existing paths.

The original filter architecture included:

• Small X-capacitor + small CM choke before the bridge rectifier,
• Large X-capacitor + large CM choke after the bridge,
• A Y-capacitor (2.2 nF) between primary and secondary ground, with a ferrite bead in series on its lead,
• A small CM choke on the output (Vo–GND).

Impedance Model and Countermeasure Evaluation

The CM loop was modeled as a voltage divider:
V_LISN = V_noise × [Z_LISN / (Z₂ + Z₃ + Z₄ + Z_LISN)]

Where:

• Z₂ = Transformer primary–secondary parasitic capacitance (~few pF → ~1 kΩ at 15 MHz),
• Z₃ = Y-capacitor + ferrite bead impedance (~1/(2πfC) + Z_bead),
• Z₄ = Output CM choke impedance (~100 Ω at 15 MHz),
• Z_LISN = 50 Ω (fixed).

Optimization levers were evaluated:

The ferrite bead had been added years earlier to suppress >100 MHz radiation from a low-inductance output choke. When the choke was upgraded to a higher-inductance, 2-turn version, the bead became redundant — and its ~20 Ω impedance at 15 MHz was actively degrading the CM filter’s high-frequency performance. Removing it restored the Y-capacitor’s full shunting capability.

Capacitor and Filter Fundamentals: Beyond “Bigger Is Better”

Capacitors are foundational to EMI filtering — but their behavior is highly frequency-dependent. The ideal impedance of a capacitor is Z = 1/(2πfC). However, real MLCCs exhibit parasitic series inductance (ESL) and resistance (ESR), leading to self-resonance.

A typical 10 nF, 0603, X7R MLCC resonates near 25 MHz. Below resonance, it behaves capacitively (Z ↓ as f ↑); above resonance, ESL dominates and Z ↑ with frequency — turning it into an inductor. This explains the rule of thumb: “large capacitors for low frequencies, small capacitors for high frequencies.” A 100 nF capacitor filters 100 kHz well but is ineffective at 30 MHz; a 100 pF capacitor is ideal at 30 MHz but useless at 100 kHz.

Effective broadband filtering uses parallel capacitor banks (e.g., 100 nF + 10 nF + 1 nF + 100 pF) or multi-stage LC filters. A single capacitor gives −20 dB/decade roll-off; an LC stage (inductor in series, capacitor in shunt) achieves −40 dB/decade — approaching ideal low-pass behavior.

Also critical: the “20 kHz PWM” label is misleading. A square wave’s harmonic content extends to f ≈ 1/(π·t_r). With a 5 ns rise time, energy spreads beyond 60 MHz. Slowing the gate driver edge with a ≥47 Ω resistor moves spectral energy downward — reducing MHz-band content — but increases switching loss. Always re-measure MOSFET junction temperature.

Systematic Countermeasure Checklist for SMPS

Focus on the three primary EMI sources: the switching MOSFET, the transformer, and the output rectifier. Coordinating fixes across all three solves >85% of cases.

MOSFET and Transformer Side

• Gate resistance: Increase to ≥47 Ω to slow t_r/t_f; verify MOSFET temperature stays within derating limits.
• Drain–source capacitance: Add a small (<220 pF), high-voltage ceramic capacitor directly across D–S if 30–80 MHz margin is insufficient — improves damping without affecting low-frequency operation.
• Ferrite beads: Place between transformer and MOSFET drain to limit di/dt; add on L/N lines or wind AC input cable 3× through a ferrite ring to suppress 40–100 MHz radiation.

Y-Capacitor Placement Strategy

• For metal-chassis products: Connect Y-capacitor between primary ground and secondary ground — provides shortest, lowest-inductance return path.
• For low-power, floating designs (e.g., integrated-MOS controllers): Connecting Y from bulk-capacitor positive rail to secondary ground often yields +3–5 dB better CM suppression than primary-ground connection — due to lower source impedance at the bulk node.
• Loop area matters: The Y-capacitor and its return trace form a magnetic loop. Keep this loop area minimal — even a 1 cm² reduction can improve 30 MHz radiation by 2–3 dB.

PCB Layout Rules That Prevent EMI at the Source

• Minimize high-di/dt loop areas: Especially the RCD clamp loop, secondary rectifier loop (the strongest electric-field radiator), and RC snubber loop.
• Shorten “hot” node traces: Long traces from MOSFET drain or rectifier anode add inductance and couple capacitively to ground planes, amplifying CM emission.
• Route with radiation in mind: Keep interface/data cables far from crystal oscillators, MCU clock lines, and fast digital signals — current radiates, not voltage. A mere 8 µA of CM current at 30 MHz produces the same 100 µV/m field at 3 m as 20 mA of DM current.

Final Engineering Principles for Reliable EMI Resolution

Before reaching for a soldering iron or changing a BOM, answer these three questions:

1. Where is the noise source? Identify the exact node with highest dV/dt or di/dt (e.g., MOSFET drain, transformer primary start, rectifier cathode). Use near-field probes or current clamps to confirm.
2. Is the mechanical/structural design appropriate? Does the metal chassis provide a continuous, low-impedance ground plane? Are heatsinks properly bonded to quiet ground? Is transformer shielding implemented and correctly grounded?
3. Is there a better way to connect the grounds? A failure that appears only with output ground tied to earth almost always indicates a CM path through the load. Model the return path as a parallel impedance network — then choose the countermeasure that delivers maximum margin per unit of cost, leakage, efficiency loss, or board area.

Every fix must be verified by measurement — and every conducted EMI correction must be followed by a radiated emissions check. As demonstrated in the 15–16 MHz case, improving conducted performance can inadvertently worsen radiated emissions if the new current path has larger loop area or higher di/dt.

FAQ