Where Is This Noise Coming From?#

The circuit should be quiet but isn’t. There’s something on the signal that shouldn’t be there — a hum, a buzz, a whine, broadband hash, or periodic spikes. Before fixing noise, identify what’s generating it and how it’s getting in. This is detective work: observe, hypothesize, test.

Characterizing the Noise#

AC-couple the scope to remove DC and focus on the noise. Set bandwidth limit to 20 MHz to reduce probe-artifact noise (unless high-frequency content is suspected). Set vertical scale to zoom in on the noise — start at 10 mV/div and adjust.

Set timebase based on suspected source:

• 5 ms/div: mains hum (50/60 Hz)
• 1 µs/div: switching converter noise (100 kHz–2 MHz)
• 100 ns/div: digital clock crosstalk (MHz range)

FFT for Frequency Identification#

Capture the noisy signal and run FFT with a Hann window. Each peak is a clue:

• 50/60 Hz and harmonics → mains-related
• Frequency matching a switching converter → switching noise
• Frequency matching a clock oscillator → digital crosstalk
• Harmonics of a known signal → distortion from nonlinearity

Compare the FFT of the noisy node to the FFT of the suspected source — if the same frequencies appear, the coupling path is identified.

Near-Field Probes for Localization#

Near-field probes (H-field loop or E-field stub) connected to the scope find which component or trace radiates noise.

H-field probe (magnetic loop): Sensitive to current loops — switching transistors, inductors, traces carrying pulsed current.

E-field probe (short stub): Sensitive to voltage nodes — high-dV/dt switching nodes, clock traces.

Move the probe slowly across the board while watching the scope. Signal amplitude increases approaching the source. Rotate the H-field probe — it’s directional (maximum when perpendicular to current flow).

Process of Elimination#

When the noise is characterized but the coupling path is uncertain:

1. Turn off suspect sources one at a time. If noise disappears when powering down the switching converter, the converter is the source
2. Disconnect cables one at a time. If noise disappears when unplugging a USB cable, that cable carries or radiates the noise
3. Move the DUT away from other equipment. If noise reduces with distance, it’s radiated; if distance doesn’t help, it’s conducted
4. Shield with grounded copper foil. If noise drops with foil between source and victim, radiated coupling is confirmed
5. Add ferrite clamp to cable. If noise drops, common-mode conducted noise on that cable is the path

Tips#

• Before blaming the circuit, disconnect the probe tip and observe noise with just ground connected — if noise persists, it’s pickup into the probe, not from the circuit
• Use spring-tip ground for clean noise measurements
• The FFT shows frequency content, not coupling mechanism — a peak at switching frequency doesn’t indicate whether noise is conducted or radiated

Caveats#

• Long ground clips act as antennas and pick up noise
• Turning things off changes circuit operating conditions — a converter that’s off isn’t producing noise but also isn’t providing power
• Multiple coupling paths can exist simultaneously — fixing one may reveal another
• Near-field measurements are qualitative — relative “more here, less there,” not calibrated dBm

In Practice#

• Noise at exactly 50/60 Hz indicates mains magnetic coupling — check for ground loops or proximity to transformers
• Noise at 100/120 Hz indicates power supply rectifier ripple — check bulk capacitor ESR
• Noise at switching frequency indicates conducted or radiated coupling from switcher — check decoupling and layout
• Noise correlated with digital bus activity indicates crosstalk — increase spacing or slow edges
• Noise that disappears when moving cables indicates the cable is part of the coupling path
• Crosstalk that appears only when two subsystems run simultaneously indicates physical coupling that the schematic treats as nonexistent — the layout made the abstraction false.
• Noise on an analog measurement that correlates with digital bus activity often shows up as periodic spikes or step changes coinciding with communication transactions — the digital IC’s supply current transients are coupling through the shared power rail into the analog circuit’s reference or supply.
• A block whose output includes oscillation in an amplifier block that isn’t supposed to oscillate is a composition failure, not a component failure. Check the feedback path, the supply bypassing, and the input/output impedance interactions before suspecting bad parts.
• Noise on an analog measurement that correlates with a digital bus’s activity is cross-subsystem coupling through power, ground, or proximity. The correlation pattern reveals the coupling mechanism: supply frequency means conducted, clock frequency means radiated, and data-dependent means ground bounce.
• Two devices that work independently but interfere with each other when both are active is frequently a blurred boundary — shared power, shared ground, or electromagnetic coupling between the devices creates an interaction path that doesn’t exist during independent testing.
• A system that works with two devices but fails when a third is added often shows up as a coordination failure — the third device introduces bus loading, supply current, or electromagnetic interference that disrupts the coordination between the first two. The third device may be functioning perfectly by its own specification.