When calculating the target impedance for inductor components in a power delivery network (PDN), the textbook formula $Z = 2\pi fL$ is only half the story. To effectively suppress noise, you must target an impedance value at least 10 times the source impedance at the specific noise frequency, while ensuring the component's Self-Resonant Frequency (SRF) sits above your noise band. Drop a standard power inductor on a 500 MHz digital rail, and its parasitic capacitance will turn it into a capacitor, passing the noise straight through. Here is the bench-tested framework for selecting, applying, and verifying inductive filtering.
The Real-World Impedance Curve: Why Ideal Inductors Fail at RF
Every physical inductor contains parasitic parallel capacitance ($C_p$) between its windings and series DC resistance (DCR). These parasitics create a Self-Resonant Frequency (SRF). Below the SRF, the component behaves inductively (impedance rises with frequency). At the SRF, impedance peaks. Above the SRF, the component behaves capacitively, and impedance plummets.
Ferrite beads are lossy inductors designed specifically to convert high-frequency noise into heat rather than reflecting it. However, they suffer from severe DC bias derating. A bead rated for 600Ω at 100 MHz might drop to 120Ω when carrying just 50% of its rated DC current. The table below maps real-world impedance values across critical frequencies to illustrate why part selection dictates signal integrity.
| Component (Manufacturer / P/N) | Type & Nominal Value | Z @ 10 MHz | Z @ 100 MHz | Z @ 500 MHz | SRF | Max DC Current |
|---|---|---|---|---|---|---|
| Wurth 742792096 | Ferrite Bead (600Ω @ 100MHz) | 120 Ω | 600 Ω (Peak) | 150 Ω | ~80 MHz | 2.0 A |
| Murata LQM2HPN1R0MG0 | Power Inductor (1.0 µH) | 62 Ω | 250 Ω | 40 Ω (Capacitive) | ~120 MHz | 2.5 A |
| Coilcraft 0402HP-10N | RF Ceramic Inductor (10 nH) | 0.6 Ω | 6 Ω | 31 Ω | ~3.5 GHz | 1.1 A |
| TDK MPZ1608S601A | Ferrite Bead (600Ω @ 100MHz) | 90 Ω | 600 Ω | 280 Ω | ~110 MHz | 1.0 A |
Identifying the Dominant Coupling Path in Your Circuit
Before throwing an inductor at a noisy rail, you must identify how the noise is actually propagating. Inductors only block conductive (galvanic) noise traveling along the copper trace. They do nothing for noise jumping through the air or across dielectrics.
- Conductive Coupling: Noise travels through the physical copper path. This is the dominant path for power rail ripple, switching regulator output noise, and ground bounce. Series inductors and ferrite beads are highly effective here.
- Capacitive Coupling: Governed by $I = C(dv/dt)$. High-speed digital edges couple into adjacent high-impedance analog traces through parasitic fringe capacitance. Inductors on the power rail won't fix this; you need increased trace spacing, guard traces, or lower $dv/dt$ slew rates.
- Radiated (Magnetic) Coupling: Governed by $V = -L(di/dt)$. High-current switching loops (like the input capacitor to a buck converter's high-side FET) act as loop antennas. This is the dominant path in the immediate near-field of a switching node.
If your noise is radiated, you might consider metallic shielding. However, shielding without proper ground-termination is useless. A floating shield merely acts as a parasitic capacitor and re-radiates the noise. Any RF shield can must be terminated to the chassis or a solid ground plane with multiple low-inductance connections (e.g., a via fence spaced at $\lambda/20$ of the highest noise frequency) to provide a return path for the induced eddy currents.
Ranked Fixes: Cost vs. Effectiveness for Noise Control
Do not treat ferrite beads or inductors as a universal cure. Placing a high-impedance bead in series with a power rail that feeds low-ESR ceramic decoupling capacitors creates an underdamped LC tank circuit. This can cause severe transient ringing and voltage overshoot when the load steps, potentially bricking a 3.3V microcontroller. Use this ranked decision tree instead.
| Rank | Mitigation Strategy | Cost | Effectiveness | Best Application |
|---|---|---|---|---|
| 1 | PCB Layout Loop Reduction | $0.00 | Critical | Minimizing $di/dt$ radiated loops at the source. |
| 2 | Local MLCC Decoupling (e.g., 100nF + 1µF) | < $0.05 | High | Providing local high-frequency charge reservoirs. |
| 3 | Series Inductor / Ferrite Bead | $0.10 - $0.50 | Medium-High | Isolating noisy sub-circuits (e.g., PLL or ADC rails). |
| 4 | Pi-Filter (CLC) or Common Mode Choke | $0.50 - $2.50 | Very High | Cleaning up dirty external DC inputs or high-speed data lines. |
The cheapest fix that actually works: Rank 1. Minimizing the physical area of your high-current switching loops on the PCB costs nothing in BOM and solves the root cause of both conducted and radiated EMI. If your input capacitor is 15mm away from your switching FET, no downstream ferrite bead will save your signal integrity. Close the loop first.
When to use Rank 3 (Target Impedance for Inductor): Use a series inductor or bead only when you need to isolate a sensitive analog block (like a 16-bit ADC or a VCO) from a noisy digital power plane. Calculate the required impedance based on the noise frequency, but always add a bulk capacitor (e.g., 10µF) before the bead and a small ceramic capacitor (e.g., 100nF) after it to dampen the LC resonance. For a deep dive into this damping behavior, the Analog Devices application note on ferrite bead LC resonance provides the exact damping resistor calculations.
Proving the Fix: Before and After Measurement Methods
You cannot manage what you do not measure. Proving that your chosen impedance for inductor filtering actually works requires rejecting the measurement noise introduced by the probe itself. Standard 10cm oscilloscope ground-spring leads act as antennas, picking up radiated noise and displaying it as power rail ripple.
Follow this numbered procedure to verify your filter with a meter or scope:
- Establish a Baseline with a Coaxial Probe: Solder a 50Ω SMA-to-pigtail coaxial cable directly across the power rail and ground plane, or use a dedicated tip-and-barrel probe (like the Tektronix TDP1500). Set the oscilloscope to 50Ω input termination and AC-coupling. This rejects common-mode radiated noise and shows true conducted ripple.
- Capture the Unfiltered FFT: Trigger on the switching frequency (e.g., 2 MHz for a buck converter). Use the scope's Fast Fourier Transform (FFT) function to identify the dominant harmonic frequencies (e.g., spikes at 50 MHz, 150 MHz, and 250 MHz). Record the peak-to-peak voltage and the dBm levels of these harmonics.
- Insert the Filter and Re-Measure: Populate your selected inductor or ferrite bead, along with the damping capacitors. Re-measure using the exact same coaxial setup and scope scale.
- Check for Ringing and DC Drop: Look at the time-domain waveform during a load transient. If you see underdamped sinusoidal ringing immediately after a step-load, your LC filter is resonating. You must either increase the ESR of the output capacitor or add a parallel damping resistor. Simultaneously, measure the DC voltage with a precision multimeter to ensure the inductor's DCR isn't causing an unacceptable voltage drop at maximum load.
- Near-Field Radiated Sweep (Optional): If your goal was to reduce radiated emissions, connect an H-field near-field probe to a spectrum analyzer. Sweep the probe over the switching node before and after filtering to confirm that the magnetic field intensity (dBµA/m) has dropped at the target frequencies.
By treating inductor impedance as a frequency-dependent, bias-dependent variable rather than a static number, and by verifying the results with proper RF probing techniques, you eliminate the guesswork from power integrity design. Always consult the manufacturer's S-parameter files and proper power-rail probing methodologies to ensure your bench measurements reflect reality.
