Guide: characterizing power supplies

What is measured, how to connect it, which instrument to use and which traps to avoid. Each section says how to do it with OSA Lab.

This guide applies to linear supplies (transformer, rectifier, LDO regulator), switch-mode supplies and DC/DC modules. Example figures are only indicative: your product specification or the component datasheets are what count.

1. Before you start

• Safety. With a mains input, use the programmable AC source or an isolation transformer and an RCD. No scope grounds on the primary side. Electronic loads have per-channel voltage and power limits too: check them first.
• Warm-up. Leave the instruments on for 30–60 min (6½-digit multimeters drift while warming up), and run the supply under test with load for a while before measuring regulation.
• Ambient temperature. Write it down; if you can, log it with one more thermocouple.
• Protection in OSA Lab. Before leaving anything running on its own, add rules in Tools › Protections: maximum heatsink temperature, minimum or maximum Vout, and "reading stops arriving". The usual action is Turn off all outputs. The manual stop is Ctrl+Shift+X (or the STOP button on the phone).

2. General setup

• Separate current and sensing. Current flows to the load through its own leads; voltage is measured with separate leads right at the output terminals (or wherever the specification says). Measuring at the load terminals includes the lead drop: 10 mΩ at 3 A is 30 mV.
• Load leads short and twisted (less inductance for the transient). Some loads have their own remote sense: use it when the load runs in CV/CR.
• One ground reference for the instruments that measure small signals (scope, analyzer). Ground loops inject 50/60 Hz and harmonics into noise measurements.

3. What to measure, and with what

4. Load regulation

What it is: how much Vout changes when the output current goes from minimum to maximum, with the input fixed.

Load regulation (%) = (Vno load − Vfull load) / Vfull load × 100
Output resistance = −ΔVout / ΔIout (slope of the line; more useful to compare designs)

1. Multimeter 4-wire at the output terminals, DC V, NPLC 10, fixed range (autorange can switch range in the middle of the sweep).
2. Load in CC. Sweep from I_min to I_max (10–20 points). Wait 2–5 s per point and average 3–5 s of readings.
3. Repeat the sweep downwards: if the down curve does not match the up curve, what you see is thermal drift (regulator or reference heating up), not regulation. To separate both effects, measure with short pulses, or with more points and less waiting.
4. On supplies with remote sense, measure with the sense connected where it will be in the application.

5. Line regulation and headroom (dropout)

What it is: how much Vout changes when the input voltage goes through its range, with a fixed load (usually full load).

Line regulation (%) = ΔVout / nominal Vout × 100 · also given in mV/V or %/V.

1. Input range: for 230 V mains, 207 to 253 V (±10 %) at least; widen it if the product must cope with more (e.g. 195–265 V). The Chroma 61602 programs it directly.
2. Fixed full load. 5–7 points, enough wait time for the input capacitors to settle.
3. Linear supplies: the critical point is the headroom at the ripple valley at minimum mains and full load. Measure the regulator input with the scope: the minimum of the waveform (valley) minus Vout must exceed the regulator dropout with margin. If the valley drops below it, 100 Hz (or 120 Hz) ripple appears at the output. With parallel LDOs (e.g. LT3045), also check current sharing at that point.
4. On switch-mode supplies, also check duty cycle and mode (PWM/PFM) at the extremes.

6. Efficiency and no-load power

η = Pout / Pin = (Vout · Iout) / real input power

• With an AC input, input power must be real power (W), not V × A (VA): a supply with transformer and capacitor has a low power factor. The Chroma 61602 measures it; a plug-in wattmeter is not accurate enough at low power.
• Efficiency vs. load curve (10 %, 25 %, 50 %, 75 %, 100 %) and no-load power: in transformer-based linear supplies, no-load power is dominated by the transformer (magnetizing current and core losses).
• Let each point settle (temperature changes efficiency).

7. Ripple with the oscilloscope

Ripple is the periodic content (100/120 Hz from rectified mains, switching frequency and harmonics, switching spikes). It is given in mV peak-to-peak and mV RMS, always with the measurement bandwidth (usually 20 MHz).

1. AC coupling, 20 MHz bandwidth limit, probe at ×1 (×10 makes probe and scope noise 10 times larger relative to the signal).
2. Spring ground (tip-and-barrel) right on the output capacitor, never the long alligator ground lead: it acts as an antenna and picks up the switching field. Better still: 50 Ω coax with a DC blocking capacitor and 50 Ω termination at the scope.
3. Check the measurement floor: with the tip touching ground at the same point (shorted probe), what you see is what your setup measures without the supply. If it looks like the measurement, the measurement is not valid.
4. Timebase: about 5 ms/div for 100 Hz ripple; a few µs/div for switching; tens of ns/div for spikes.
5. Take several captures: peak-to-peak is statistical (the more captures, the larger). Report the worst of N captures.

8. Noise with the FFT analyzer (35670A)

With low-noise regulators (in the µV RMS range) the oscilloscope is useless: its own noise is hundreds of µV or mV. Measure the noise spectral density in V/√Hz and integrate it over the band given in the datasheet (usual for LDOs: 10 Hz – 100 kHz). As a reference, the LT3045 specifies about 0.8 µV RMS over 10 Hz – 100 kHz and about 2 nV/√Hz at 10 kHz.

Setup

• Low-noise preamplifier (LNA) between the supply and the 35670A, battery powered. The analyzer's own noise (measure it, see below) (specified < −140 dBV/√Hz ≈ 100 nV/√Hz above 1.28 kHz on the most sensitive range, worse below) is about 50 times that of a 2 nV/√Hz LDO: without an LNA you are measuring the analyzer. Two conditions: about 60 dB of gain (×1000, so the analyzer floor referred to the input is ~0.1 nV/√Hz) and LNA self-noise below ~1 nV/√Hz (otherwise you measure the LNA). An SR560 (4 nV/√Hz) is not enough for an LT3045; you need a design like the one in Analog Devices note AN-159 (≈0.5 nV/√Hz, 80 dB).
• DC block at the LNA input (film or low-leakage electrolytic capacitor + resistor to ground). The time constant sets the corner: τ = R·C; f_c = 1/(2πRC). To measure from 10 Hz with margin, f_c of 0.1–1 Hz. Let the capacitor charge before measuring (several τ).
• Closed metal box for the supply and the LNA. Mains (50/60 Hz and harmonics) gets in everywhere: if possible, power the DUT input from batteries or a clean linear supply, and use the 35670A floating input to break ground loops.
• Short coax cables. No phones or switch-mode supplies nearby.

Procedure

1. Noise floor. Short the LNA input (inside the box) and measure. That is the limit of your system. Press "Use last measurement as noise floor".
2. Measurement. Connect the supply, same settings, and measure. Enable "Subtract noise floor" (subtracted in quadrature: √(S² − S_floor²)). It is only reliable when the measurement is clearly above the floor (×2 or more). If they are close, the result is "noise is below the floor", not a number.
3. Log sweep (3 spans of 1600 lines: 400 Hz, 6.4 kHz and 102.4 kHz): good low-frequency resolution and reaches 100 kHz. A single 102.4 kHz span with 1600 lines has 64 Hz resolution, so integrating from 10 Hz is not possible.
4. RMS averaging: 50–100. Hanning window for noise (Flat top only to measure the amplitude of a tone).
5. LNA gain: enter it in the panel; measure it first with a known sine (the 35670A has an internal source) through an attenuator.
6. Read the integrated noise (10 Hz – 100 kHz and sub-bands) and the spot noise at 10 Hz, 100 Hz, 1 kHz, 10 kHz and 100 kHz.

Cross-correlation (two preamplifiers)

To go below your own preamplifier's noise: two identical LNAs connected to the same point, each to one 35670A channel, with separate supplies. The cross spectrum CH1×CH2 is averaged: the supply's noise is common to both channels and stays; each LNA's noise is uncorrelated and its residue falls as √N (in power). With 1000 averages the residue drops about 5.6 times in V/√Hz relative to one LNA's noise; with 10000, 10 times. The panel draws that limit (dashed red) and says how much of each band is above it with margin. With two channels the maximum span is 51.2 kHz. Averages take time: 1000 averages of the 400 Hz span with 1600 lines take ~67 minutes; start with 200–500 and watch the limit.

9. Load transient

How the output responds to a current step: maximum dip/overshoot (mV) and recovery time until it stays within a band (e.g. ±1 % or ±10 mV). It tells you about loop stability: lots of ringing = little phase margin.

1. Load in dynamic mode: levels (e.g. 10 %↔100 % or 50 %↔100 % of rated current), frequency (100 Hz–1 kHz to see the whole step) and a defined slew rate (A/µs). Slew rate changes the result a lot: always state it. At high slew rates, load lead inductance is the limit: short twisted leads.
2. Scope at the output, AC, 20 MHz, ×1 probe with spring ground. If the load has a sync output, trigger on it; otherwise trigger on the output edge.
3. Look at both edges: load step up (voltage dip) and down (overshoot).
4. With paralleled supplies or LDOs, the transient also shows whether they share current stably.

10. Power supply rejection (PSRR)

How much the regulator attenuates a disturbance at its input vs. frequency: PSRR (dB) = 20·log(Vin,ac / Vout,ac).

• A small AC signal has to be added to the DC input: injection transformer (line injector), a power amplifier, or a modulated pre-regulator. Small amplitude (tens to hundreds of mV) to stay linear and within the dropout headroom.
• Input and output are measured at the same time. With the 35670A in two channels (frequency response, internal source) high PSRR can be reached (a low-noise LDO can exceed 70–80 dB at low frequency). With generator + scope (OSA Lab Bode plot) the limit is around 50–60 dB because of scope resolution.

11. Thermal test

1. K thermocouples on the hot spots: regulator (tab or package), heatsink, transformer (winding and core), rectifier, electrolytic capacitors (their life depends on temperature) and one in the ambient air.
2. Worst case: maximum mains and full load (more dissipation in linear supplies), inside the final enclosure, without extra ventilation.
3. Hold until stable: less than 1 °C change in 10–15 min (large transformers need more than an hour).
4. Report the rise over ambient (ΔT) as well as the absolute temperature, and extrapolate to the maximum operating ambient.

12. Start-up, protection and drift

• Start-up: scope in single-shot on the Vout rising edge when the input is switched on: rise time, overshoot, monotonic start-up (no steps) with minimum and maximum load, and with capacitive load if the application has it.
• Current limit and short circuit: load in CV lowering the voltage, or in short mode; measure the limit current, whether it is foldback and whether it recovers on its own when the short is removed. Watch temperatures.
• Output voltage accuracy and its spread between units (several boards).
• Long-term drift: 8–24 h log with the multimeter (NPLC 10) and ambient temperature: OSA Lab records the whole session, and the plot cursors give the drift per minute between two instants.

13. Report and checklist

OSA Lab characterization produces the PDF with summary, setup, plots and table, and the CSV with every point (folder OSALab/caracterizacion). The 35670A panel and the ripple tool export their curves to CSV.

© 2026 OSA Electronics · OSA Lab