Measuring supply ripple: the ground clip is the antenna
A 3-inch probe ground lead turns a switching spike into ringing that is not on the rail. Tip-and-barrel, coax, bandwidth limits, and the null test.
Most reported switching-regulator ripple is partly measurement artefact, and the usual culprit is the few inches of wire hanging off the probe. Before adding capacitance to fix ripple, prove the ripple is real: connect the probe tip and its barrel directly across the output capacitor, band-limit the scope to the frequencies the rail actually carries, then ground the probe tip to the same point and confirm the display goes quiet. If it does not, everything measured before that moment was the loop, not the rail.
That is the whole answer. The rest is why it happens and how to do each step.
The ground lead is a loop, and a loop is an antenna
A standard 10× probe ships with a flying ground lead and an alligator clip. Used as supplied, the tip and that lead enclose a couple of square centimetres — roughly thirty times what a tip-and-barrel contact encloses, for the lead geometry Fig 1 assumes. A switching converter puts amps through its hot loop with edges of a few nanoseconds, radiating a magnetic field that the probe loop sits inside. Any changing field through a loop induces a voltage in it, and the scope cannot tell that voltage apart from the one on the rail.
Analog Devices’ AN101 shows the size of the error directly. Its Figure C1 is a switching spike measured through a continuous coaxial path — the body of the spike well defined, the disturbance after it contained. Figure C2 is the same event with only one thing changed:
Figure C2 depicts the same event with a 3 inch ground lead connecting the coaxial shield to the circuit board ground plane. Pronounced signal distortion and ringing occur.
Three inches. Not a long lead by bench standards — shorter than the one on most probes — and it converts a defined spike into ringing that does not exist on the rail. Both photographs were taken at 0.01 V/division, so this is not a sub-microvolt curiosity; it is the scale at which people judge whether a regulator is acceptable.
Tip and barrel, which costs nothing
The fix that requires no parts is to stop using the ground lead. Pull the plastic tip cover and the ground clip off the probe, and press the exposed tip onto the capacitor’s positive terminal with the grounded barrel against its return terminal. The enclosed area falls from square centimetres to roughly the diameter of the probe barrel. The probe loading calculator gives the ring frequency of the ground lead just removed, against the bandwidth of the scope.
TI’s SLYT740 puts it as plainly as an application note can:
When measuring the switch node or output of a DC/DC converter, it is bad practice to use the alligator clamp for ground on the oscilloscope probe. Instead, take the measurement using the tip-and-barrel technique, which will provide a more accurate waveform.
Two details decide whether it works. Measure across the output capacitor’s own
terminals, not at a convenient via or the far end of the load: ripple is a
voltage that exists between two specific points, and every millimetre of copper
between the capacitor and the probe adds inductance whose L·di/dt joins the
reading. And keep the barrel contact as short as the tip contact, since a long
return defeats the point of shortening the lead.
Coax beats tip and barrel, when the amplitude is small
Below a millivolt or so, even a tip-and-barrel 10× probe struggles. SLYT740 reports better results again from soldering a coaxial connector to the output and running it straight to the scope:
The advantage of this technique is that the measurement is 1-to-1 ratio, compared to 10-to-1 for a typical probe. With a standard 10-to-1 probe, the noise in the measurement is amplified 10 times.
The 10× attenuator divides the signal by ten before the scope’s own input noise is added, so the noise floor referred to the rail is ten times worse. A 1× coaxial path removes both that penalty and the probe’s loop. AN101 goes further for its sub-millivolt work, keeping a “strictly maintained coaxial environment” all the way — including the AC coupling capacitor — from the regulator through a 40 dB preamplifier to the oscilloscope.
The practical version for a board you are designing: put an SMA or a U.FL footprint across the output capacitor. It costs a few cents and one square millimetre, and it converts every future ripple argument from opinion to measurement.
Ripple and spikes are two different problems at two different frequencies
Deciding what bandwidth to use requires knowing what is being looked for. AN101 separates the rail’s content into two populations:
RIPPLE: TYPICALLY 100kHz to 3MHz SWITCHING SPIKES: HARMONIC CONTENT APPROACHING 100MHz
These want opposite settings. Ripple is the sawtooth at the switching frequency, and it is what the buck ripple calculator predicts from inductance, capacitance and ESR. Spikes are the nanosecond edges of the switch node coupling through parasitics, and no capacitance value in the ripple formula predicts them — they answer to layout and to the capacitor’s ESL.
So measure twice. With the scope’s 20 MHz limit engaged, the sawtooth is visible and comparable with the calculation. With full bandwidth, the spikes appear — and so does every artefact the probe is capable of inventing, which is exactly why the probing has to be right before the number means anything.
Quoting a single ripple figure without stating the measurement bandwidth is meaningless, and it is why two engineers measuring the same rail routinely disagree by a factor of three.
The null test: ground the tip and look again
The step that turns a measurement into evidence takes ten seconds. AN101 describes it as the way to finish the job:
The measurement is completed by verifying that indicated results are not corrupted by common mode components or ground loops. This is done by grounding the oscilloscope input near the measurement point. Ideally, no signal should appear.
Touch the probe tip to the same ground the barrel is on, right at the capacitor, and look at the screen on the same volts-per-division as the real measurement. Whatever remains is what the probe and the scope are manufacturing by themselves. If that residue is comparable to the ripple just measured, the ripple figure is not yet a fact.
This test costs nothing and settles most bench disputes about power supplies outright.
What good practice looks like, in order
- Fix the layout first. AN101 is emphatic that measurement follows design: “If the circuit is sound, the board layout proper and appropriate components used, then, and only then, may meaningful measurement proceed.” A probe cannot rescue a bad hot loop.
- Probe across the output capacitor terminals, tip and barrel, no ground lead.
- Set 20 MHz bandwidth limit and record the ripple. Compare it against the buck ripple calculator.
- Go to full bandwidth and record the spikes separately.
- Run the null test. Ground the tip; confirm the display goes quiet.
- State the bandwidth alongside any number reported to anyone else.
What this changes about fixing ripple
Once the measurement is trustworthy, the diagnosis usually changes. Ripple at the switching frequency that matches the calculation is a component problem, and the calculator says which component: in the resistive regime the answer is capacitor ESR and adding capacitance does nothing at all, which is where the buck ripple comes from.
High-frequency spikes that survive a correct measurement are a layout problem — the hot loop and the capacitor’s ESL — not a capacitance problem. And ringing that vanishes when the ground lead comes off was never on the board.
A related trap sits one step downstream. Reaching for a ferrite bead to clean up what the scope shows can make the rail genuinely worse, because a bead and a low-ESR ceramic form an undamped LC that amplifies near its resonance; the ferrite bead filter calculator shows where that peak lands, and it is often squarely on the switching frequency being chased.
Common mistakes
- Judging a regulator with the ground clip on. The first measurement of any new board is worth repeating tip-and-barrel before anyone reacts to it.
- Comparing a full-bandwidth number to a datasheet’s 20 MHz number. They are measurements of different quantities.
- Probing at a convenient point. Ripple is a voltage between two terminals; measured anywhere else it includes the copper in between.
- Skipping the null test and then arguing about the result for an afternoon.
- Adding output capacitance because the screen looks bad, before knowing whether the screen is showing the rail or the probe.
Sources
- ADI AN101, Minimizing Switching Regulator Residue in Linear Regulator Outputs — Jim Williams. Appendix C, “Probing Technique for Sub-Millivolt, Wideband Signal Integrity”, is the source of the 3-inch ground lead comparison and the null test.
- TI SLYT740, Reducing noise on the output of a switching regulator — the tip-and-barrel and coaxial recommendations, measured on a real converter.
- TI SLVA630, Output Ripple Voltage for Buck Switching Regulator — what the ripple should be, once it can be measured.