HOW-TO GUIDE
What Oscilloscope Bandwidth Do You Need for Switching Power Supplies?
By the OhmSavvy Research Team ·
Safety first: the primary side of a switching supply is at mains potential
- On an offline supply (one that plugs into the wall), everything before the transformer, including the main switching transistor, is connected to mains. A bench scope's probe ground clip is tied to mains earth, so clipping it there shorts that node to earth and can destroy the probe, the scope or the supply, or injure you.
- Never float the scope by defeating its earth connection. Use a differential probe rated for the voltage, or power the device under test through an isolation transformer.
- The large input capacitor can hold a dangerous charge after unplugging. Treat it as live until you have confirmed it is discharged.
The short answer
Two jobs, two bandwidth figures
Switching frequency vs edge speed
Switching frequencies are low by oscilloscope standards, commonly somewhere from tens of kilohertz to a few megahertz depending on the design. If the switch node were a sine wave, a few megahertz of bandwidth would do. It isn't: it is a square-ish wave whose edges are as fast as the transistor can switch, often tens of nanoseconds and, in some fast designs, only a few. Those edges contain frequency content far above the switching frequency, the same reason a square wave needs more bandwidth than a sine of the same frequency.
For the measurement itself, see how to measure a switching power supply waveform with an oscilloscope.
So the first question is not "what frequency does it switch at?" but "what do I want to see?" Checking that the converter switches at all, measuring its frequency or its duty cycle, needs little bandwidth. Measuring the edges, their overshoot and ringing needs a lot more.
The 0.35/rise-time rule and the 3-5x margin
For most scopes below about 1GHz, bandwidth and rise time (10% to 90%) are linked by bandwidth ≈ 0.35 ÷ rise time. A scope whose own rise time equals your edge would show that edge about 41% slower than it really is, because the displayed rise time is roughly the root-sum-of-squares of the two (√2 ≈ 1.41). The usual guideline is a scope rise time 3-5x faster than the edge, which means 3-5x the bandwidth the 0.35 rule gives. The full derivation is in what is oscilloscope bandwidth; the steps for a power supply are:
01
Find the edge
Fastest rise or fall time you need to see
02
0.35 ÷ tr
Bandwidth matching that edge
03
× 3 to 5
Margin for a small error
04
Check the probe
Probe rating caps the system
If you don't know the edge speed, the transistor's datasheet gives typical rise and fall times under its test conditions. The real circuit is usually slower, because gate drive strength and layout slow it down, so the datasheet figure is a sensible worst case.
Worked examples
Applying the rule to a range of edge speeds. All figures follow from bandwidth = 0.35 ÷ rise time:
| Edge rise time | Bandwidth matching the edge (0.35 ÷ tr) | With a 3-5x margin |
|---|---|---|
| 100 ns | 3.5MHz | 10.5-17.5MHz |
| 50 ns | 7MHz | 21-35MHz |
| 20 ns | 17.5MHz | 52.5-87.5MHz |
| 10 ns | 35MHz | 105-175MHz |
| 5 ns | 70MHz | 210-350MHz |
The same numbers at the low end of the margin (3x), on a logarithmic scale. A 50MHz scope meets the 3x margin for edges down to about 21 ns (between the 50 ns and 20 ns rows), while a 100MHz scope meets it down to about 10.5 ns, just short of the 10 ns row:
Bars use a logarithmic scale: each step is ten times the last.
Checking the 10.6 ns figure: a 100MHz scope has a rise time of 0.35 ÷ 100MHz = 3.5 ns, and √(10² + 3.5²) = √112.25 ≈ 10.6 ns, about 6% slow. On a 50MHz scope (7 ns) the same edge shows as √(10² + 7²) = √149 ≈ 12.2 ns, about 22% slow. Both are usable for seeing the edge; only the first is close enough to quote as a number.
A full example: a 500kHz buck converter
Say you are debugging a 12 V to 5 V buck converter that switches at 500kHz, and you expect switch-node edges of about 15 ns. The switching frequency alone suggests only 1.5-2.5MHz (3-5x of 500kHz), which is enough to confirm the frequency and duty cycle. The edges tell a different story: 0.35 ÷ 15 ns ≈ 23.3MHz, and 3-5x that is about 70-117MHz. A 100MHz scope sits inside that range; a 50MHz scope (7 ns) would show the 15 ns edge as roughly √(15² + 7²) = √274 ≈ 16.6 ns, about 10% slow. That is fine for spotting whether the edge has excessive overshoot, less so for quoting a precise rise time.
Overshoot is often the reason to look closely at edges in the first place. The peak of the ringing, not the supply voltage, is what the switching transistor has to withstand, so an under-bandwidth scope that rounds off the peak can make a marginal design look safe. When the edge is the point of the measurement, err toward the 5x end of the margin.
Why ripple and switching-edge measurements need different bandwidth
Output ripple is a small, relatively slow signal: it repeats at the switching frequency and its main shape is a gentle triangle. Extra bandwidth adds nothing to that figure except noise, and many power supply datasheets quote ripple and noise measured with a 20MHz limit. Switching edges are the reverse: the information is in the fast content, so any limit hides it.
| Measurement | What sets the bandwidth | Typical setting |
|---|---|---|
| Output ripple (mV p-p) | Matching the datasheet test method | 20MHz bandwidth limit on, AC coupling |
| Switching frequency and duty cycle | The switching frequency itself | Any bench scope; full bandwidth |
| Switch-node edges, overshoot, ringing | Edge rise time (0.35 ÷ tr × 3-5) | Full bandwidth, limit off |
| Gate drive waveform | Gate rise time, usually similar to the switch edge | Full bandwidth, limit off |
| Start-up and load transients | Usually slow: microseconds to milliseconds | Modest bandwidth; Single trigger mode |
The full ripple procedure, including AC coupling and V/div settings, is in how to measure ripple voltage with an oscilloscope. For one-off events like start-up, the trigger guide covers Single mode.
Probe and ground lead limits
The bandwidth printed on the scope only applies if the rest of the chain keeps up. On a switching supply, the probe is usually the weak point:
Probe rating
A probe rated below the scope caps the whole system at the probe's figure.
1x mode
A passive probe in 1x mode usually has far less bandwidth than in 10x. Use 10x for edges.
Long ground lead
Its inductance rings with the probe capacitance on fast edges, adding false ringing.
Ground spring
A short spring at the switch node's ground keeps the loop small and the edge honest.
Differential probes
Each has its own bandwidth rating, which may be lower than the scope's. Check it.
Compensation
An uncompensated 10x probe distorts edges at any bandwidth. Adjust it first.
Compensation takes a couple of minutes with the scope's built-in square wave; see how to compensate an oscilloscope probe.
Isolation and mains safety on the primary side
Most hobby measurements are on the secondary side of an isolated supply or on a low-voltage DC-DC converter, where the output ground is a safe reference for the probe ground clip. The primary side of an offline supply, and the high-side switch node of a half-bridge, are not referenced to earth, so a grounded probe cannot be clipped across them.
- Differential probe: measures between two points without connecting either to earth. Check its voltage rating and bandwidth against the job.
- Isolation transformer on the device under test: removes the earth reference from the supply being tested, not from the scope. The scope stays earthed.
- Isolated-channel instruments: scopes with isolated inputs exist for this work; check that the rating covers your voltage.
If you only need the mains voltage, a properly rated multimeter is the simpler tool; see oscilloscope vs multimeter.
Which scopes in our reviews fit
The Rigol DS1054Z (50MHz, 7 ns rise time by the 0.35 rule) handles ripple, switching frequency and duty cycle easily, and shows edges of roughly 21-35 ns or slower with little error. The Siglent SDS1104X-E and Hantek DSO2D10 (both 100MHz, 3.5 ns) halve that to roughly 10.5-17.5 ns. Four channels on the Rigol and Siglent let you watch gate drive, switch node, input and output together; the Hantek has 2. The pocket FNIRSI DSO152 (200kHz) is too slow for switching supplies: a 100kHz converter already sits at half its bandwidth. Compare them all in the best beginner oscilloscopes roundup or the oscilloscope buying guide.
Bench scopes for power supply work
50MHz, 4 channels, 1 GSa/s. Bandwidth is modest for RF work but fine for typical hobby supplies.
100MHz, 4 channels, 1 GSa/s: more bandwidth headroom for faster edges.
100MHz, 2 channels, 1 GSa/s, with a built-in waveform generator in a compact unit.
As an Amazon Associate I earn from qualifying purchases. Prices and availability are shown on Amazon.
Frequently asked questions
Is a 50MHz oscilloscope enough for switching power supplies?
For most hobby work, yes. A 50MHz scope has a rise time of about 7 ns, so by the 3-5x guideline it shows edges of roughly 21-35 ns or slower with little error, and it is more than enough for output ripple, which is usually measured with a 20MHz limit. Very fast edges will look slower than they really are.
Why does my switch node show ringing that the circuit may not have?
Often the probe's long ground lead is the cause. Its inductance forms a resonant circuit with the probe's input capacitance, and a fast edge makes it ring. Swap the ground lead for a short ground spring and measure again. Ringing that shrinks or changes frequency was coming from the probe setup, while ringing that stays the same is more likely real.
Do I need a differential probe for a switching power supply?
For anything on the mains side of an offline supply, or any node that is not referenced to earth, such as a high-side switch node, a rated differential probe is the safe choice. For the low-voltage output of an isolated supply, a standard passive probe with its ground on the output ground is normally fine.
Does sample rate matter as much as bandwidth for power supplies?
It matters too. A common guideline is a sample rate several times the bandwidth, so the scope has enough points to draw a fast edge. A 1 GSa/s scope takes one sample per nanosecond, which is adequate for edges of tens of nanoseconds. Sample rate can drop when several channels share it or at slow timebases.
What bandwidth do I need for GaN or SiC power supplies?
More than for most silicon designs. GaN and SiC transistors can switch in a few nanoseconds or faster, so the 0.35/rise-time rule with a 3-5x margin points to several hundred megahertz or more for a faithful edge. Check the transistor datasheet for its typical rise and fall times. Output ripple still uses the 20MHz limit, so a modest scope stays useful for that half of the job.
