HighSNR Lab · Application Note AN-001 · Public Release

Frequency-Dependent Distortion in Class II Ceramic Capacitors

Why THD peaks near the RC transition band—and how an incorrect C(V) update can create false harmonics.

A concise engineering note for analog designers, signal-chain engineers, and model developers working with X7R and X5R MLCCs.

Boris KuznetsovJuly 27, 2026HighSNR Lab AN-001

Engineering takeaways

Three checks before trusting an MLCC distortion result

01

The transition band is the critical region

In a nonlinear RC low-pass network, capacitor voltage swing and circuit current are simultaneously significant near cutoff. That is where output THD can peak.

02

C(V) is not enough for transient simulation

A locally frozen capacitance update can violate the charge relation. A credible nonlinear model should be formulated and checked through Q(V).

03

Convergence is part of the model

If a simulated harmonic result moves when the time step is reduced, it is a numerical diagnostic—not yet a dielectric prediction.

Controlled simulation

THD rises toward the RC transition region, then falls

The charge-consistent model captures the physical interaction between nonlinear capacitor current, resistor voltage drop, and output attenuation. The comparison model is deliberately coarse and is included as a numerical counterexample.

  • Controlled symmetric Cdiff(V) model
  • Explicit Q(V) formulation
  • FFT-based harmonic extraction
  • Reproducible Python source
THD versus frequency for charge-consistent and naive nonlinear capacitor models
Controlled example: R = 1 kΩ, C0 = 1 µF, reference cutoff 159.2 Hz. Numerical values are not a universal prediction for a commercial MLCC.
Normalized capacitance versus DC bias for Murata GRM155R60J105KE19
Real-component context from a Murata SimSurfing export: GRM155R60J105KE19, 1 µF, 6.3 V, X5R, 0402; header conditions 25 °C and 10 mV RMS.

Real-component context

A real X5R curve shows why operating point matters

AN-001 keeps its main simulation deliberately controlled, then adds a separate vendor-data case to show realistic DC-bias curvature. The vendor curve is not presented as a measured large-signal THD model.

Source limitations are disclosed: the export does not record measurement frequency, dwell time, sweep direction, or uncertainty.

Independent engineering review

Need to trust a nonlinear model before the board is built?

HighSNR Lab reviews component models, operating assumptions, convergence, and distortion risk in precision analog signal paths.