Conducted Emission Analysis

Yes. It refers to the noise current that leaks out externally through the power cables or signal cables of electronic equipment. It is generally measured in the frequency band of 150kHz to 30MHz and evaluated using a standardized impedance network called an LISN (Line Impedance Stabilization Network).

Below 150kHz is too close to the fundamental operating frequency of the power supply, making separation difficult, and above 30MHz, cables start to behave as antennas, entering the realm of radiated emission (RE). In other words, conducted EMI and radiated EMI are separated by frequency.

Almost all products equipped with switching power supplies. For example, in an EV's onboard charger (OBC), a DC-DC converter (48V→400V) operates at a switching frequency around 100kHz, so its harmonics directly overlap with the 150kHz band. The design goal is typically to pass CISPR 25 Class 5 with a 6dB margin.

An LISN (Line Impedance Stabilization Network) has two roles. One is to stabilize the power supply impedance seen from the DUT (Device Under Test) to 50Ω in the measurement frequency band. The other is to block noise from the external power supply to stabilize the measurement.

The standard LISN circuit specified in CISPR 16-1-2 terminates the power line with a 50μH inductor and connects it to the 50Ω input of a spectrum analyzer via a coupling capacitor (0.1μF to 1μF). Above 150kHz, the impedance of the 50μH inductor becomes sufficiently high, and the impedance seen from the DUT side is dominated by 50Ω.

The basic approach is to use a SPICE model as an L-R parallel circuit with a DC-blocking coupling capacitor (0.1μF) in series. However, for high-precision analysis, it is necessary to include parasitic components of the LISN (self-resonant frequency, connector contact resistance, ground impedance) in the model. The best practice is to obtain S-parameters from actual measurements and create a broadband model.

Let the noise current flowing on the positive side of the power line be $I_1$, and on the negative side be $I_2$. Then, the differential mode (DM) current and common mode (CM) current can be separated as follows:

DM noise is primarily generated by the current ripple caused by the switching device's ON/OFF and flows in opposite directions on the power supply's outgoing and return paths. On the other hand, CM noise is the current that leaks to earth via the stray capacitance between the switching node and the chassis, flowing in the same direction on both the outgoing and return paths.

For example, when a MOSFET switches in an automotive DC-DC converter, the drain voltage changes steeply (e.g., 0V→400V) (dv/dt = tens of V/ns). This dv/dt generates CM noise through the stray capacitance (tens of pF) between the MOSFET package and the heat sink. It can be estimated by $I_\text{CM} = C_\text{stray} \cdot \frac{dv}{dt}$. For 10pF and 20V/ns, that's an instantaneous leakage current of 200mA.

There are two main types of noise source models for switching power supplies: the voltage source model and the current source model. DM noise can be represented as an equivalent current source from the FFT of the trapezoidal switching current. The harmonic envelope of a trapezoidal wave is:

It's important to note that the trapezoidal wave envelope has two breakpoints (corner frequencies). At the first breakpoint $f_1 = 1/(\pi t_\text{on})$, the spectrum begins to attenuate at -20dB/dec, and at the second breakpoint $f_2 = 1/(\pi t_r)$, it changes to -40dB/dec. Faster rise times increase high-frequency components, so special care is needed, especially with SiC/GaN devices.

CM noise is determined by the dv/dt of the switching node and stray capacitance, so an equivalent voltage source model is used. It is modeled as a Thévenin equivalent circuit, with the switching node voltage $V_\text{sw}(t)$ as the voltage source and parasitic impedance in series.

Insertion loss is defined as the ratio of the LISN terminal voltage before and after inserting the filter:

The design goal is as follows. First, add a margin to the difference between the standard's limit value $L_\text{limit}(f)$ and the measured (or simulated) value $V_\text{meas}(f)$:

An LC filter with one stage (L-C) provides -40dB/dec attenuation. For frequencies above the cutoff frequency $f_c$:

For example, if 40dB attenuation is needed at 150kHz, setting $f_c$ to about 15kHz makes it achievable with a single-stage π-type LC filter. However, in reality, the filter's high-frequency characteristics degrade due to the inductor's self-resonant frequency (SRF) and the capacitor's ESL, so component selection becomes very important.

The most basic approach is SPICE circuit simulation. Connect the three elements—the switching model of the power conversion circuit, the LISN equivalent circuit, and the EMI filter—run a time-domain analysis (Transient), perform an FFT on the LISN terminal voltage waveform, and output the spectrum.

There are three important points. (1) Window function: Since switching power supplies are periodic signals, always cut the data at integer multiples of the switching period. Using a non-periodic window causes spectral leakage. (2) Time resolution: To evaluate up to 30MHz, at least 60MHz sampling (≈16.7ns step) is required from the Nyquist theorem. Practically, 100MHz or higher is desirable. (3) Confirming steady state: Use data after the transient response has sufficiently converged. Including data immediately after startup will not yield the correct steady-state spectrum.

Good question. The accuracy of CE analysis is almost entirely determined by the accuracy of the parasitic parameters. These four are particularly important:

Ignoring a capacitor's ESL (Equivalent Series Inductance) is particularly fatal. For example, if the ESL of a 0.1μF MLCC (multilayer ceramic capacitor) is 1nH, its self-resonant frequency is:

Above the SRF, the capacitor behaves as an inductor rather than a capacitor. This means the filter capacitor becomes ineffective above 16MHz. This is the cause of bugs that remain invisible when using ideal capacitor models in SPICE.

Let's summarize the key points for choosing.