Power Supply Ripple and Noise: How to Measure, Read the Limits, and Spec for Reliability
Ripple and noise is the spec most engineers misread and most scopes mis-measure. A datasheet says “≤50 mVpp”; your bench shows 150 mVpp. Usually the supply is fine and the measurement is wrong. This guide is for the person qualifying a unit before it goes on a BOM—exactly the “how to measure power supply ripple and noise testing” job.
What ripple and noise are
Ripple is the AC component at the switching frequency (50–150 kHz typical), riding on the DC output. Noise is the broadband spike from fast edges and parasitics, up to tens of MHz. Together they are “ripple and noise,” almost always quoted in mVpp (peak-to-peak). A 24V/10A rail at 50 mVpp is 0.2%—tight. A 5V logic rail at 50 mVpp is 1%—loose.
The 20 MHz bandwidth limit is not optional
Counterintuitive point one: the number on the datasheet is measured with a 20 MHz bandwidth limit. Your 200 MHz scope, used wide open, shows 2–3× the real ripple because it picks up every edge from the switcher and the lab. Turn on the 20 MHz limit, and the 150 mVpp collapses to ~50 mVpp. The supply did not change; your measurement did. Always measure ripple with the limit on, or you will reject good parts.
Probe technique beats probe price
A 10× passive probe with its ground lead acts as a loop antenna—the longer the lead, the more noise you invent. The right method is a coax tip or a short ground spring; keep the loop under 2 cm. A Keysight N2820A current probe or a Rigol with a spring ground turns a garbage reading into a real one. Most “failures” we see in incoming inspection are 6-inch ground clips, not bad supplies.
Lower ripple can hide a slower supply
Counterintuitive point two: a cheaper supply with a lower ripple number is not automatically better. Heavy filtering that crushes ripple also slows transient response—when your load steps, the output sags before recovery. A medical monitor or a servo rail cares more about transient recovery than about 10 mV less ripple. For “medical power supply requirements,” IEC 60601-1 sets leakage and isolation; ripple matters only where it couples into a sensing front-end. Spec the number your load cares about, not the smallest.
Ripple, heat, and MTBF
Ripple heats the output caps through ESR loss: P = I² × ESR at the ripple frequency. At 10A and 20 mΩ ESR, that is 2W of pure loss in the cap—enough to age it. MTBF figures (often “100,000 hours”) are calculated at 25 °C per MIL-HDBK-217F or Telcordia; at 60 °C real life is about one quarter. A supply with low ripple runs its caps cooler and quietly lives longer, which is why the review page’s “reliability” claim should point here, not at a sticker.
A measurement checklist
- Set scope to 20 MHz bandwidth limit; DC coupling, AC math off.
- Short ground spring or coax tip; loop <2 cm.
- Load to the real operating point (40–60%, not full).
- Measure at the output terminal, not the cable end.
- Record both mVpp and the dominant frequency.
- Repeat at 60 °C if the unit ships hot.
Limits by application
- Logic / ADC front-end: <20 mVpp, low broadband noise.
- LED driver: <100 mVpp is fine; LEDs do not read ripple.
- Motor / industrial: <50 mVpp, fast transient recovery.
- Medical sensing: leakage and isolation first; ripple only where it couples in.
Closing
Ripple and noise is where a supply either proves itself or gets blamed wrongly. Measure it the way the datasheet does—20 MHz limit, short ground, real load—and you stop rejecting good units and start qualifying the right ones. The page that teaches the measurement out-ranks the one that just quotes a number, because the engineer who got burned once searches for “how to measure” before “how good.”
