Oscilloscope Probing

note

2026.09.12

How to get a true waveform out of an oscilloscope: probe loading, 10:1 and 1:1 passive probes, low-impedance and active probes, differential and isolated probes, the ground lead and its resonance, analog bandwidth versus sample rate, the bandwidth limit, vertical resolution and how to use it, coupling and termination, ripple and switch-node and gate-drive measurements, current probes, and the floating-scope hazard.

Scope: measurements with a digital oscilloscope on power electronics, mixed-signal, and general analog boards. The probe, the scope input, and the acquisition settings are treated as one measurement chain. Instrument-specific menus are not covered; the settings named here exist on every scope under some name.

Common errors

Each of these produces a waveform that looks plausible and is wrong.

Long ground lead on a fast edge. The 15 cm alligator lead and the probe capacitance resonate near 100 MHz. The ringing on the screen belongs to the probe, not the circuit. Use a spring ground tip at the node.
Bandwidth limit left on when looking at switching edges, ringing, or gate drive. A 20 MHz filter turns a 5 ns edge into a 17 ns edge and hides overshoot completely.
Bandwidth limit left off when measuring ripple to a specification that is written at 20 MHz, or when a small signal sits under wideband noise. The trace is five times noisier than it needs to be.
Reading the scope's bandwidth as its sample rate, or the reverse. Bandwidth is the analog front end. Sample rate is the digitizer, and it drops at slow timebases until the record aliases.
Signal occupying two divisions of an eight-division screen. Three quarters of the converter's codes are unused. Ripple, overshoot, and noise measurements come out coarse.
Wrong probe attenuation setting. A 10:1 probe on a channel set to 1:1 reads ten times low. Many probes signal the ratio through a pin on the BNC; many do not.
Uncompensated 10:1 probe. Square waves tilt, amplitudes are wrong above a few kilohertz, and nothing on the scope indicates it.
A 10:1 probe for millivolt ripple. The scope's own noise is multiplied by ten at the tip. A 1:1 probe or a terminated coax has a ten times lower noise floor.
Ground clip on a node that is not ground. A single-ended probe's ground lead is connected to protective earth through the scope. Clipping it to a high-side source or to the primary return of an offline supply shorts that node to earth.
Probe loading changes the circuit. A 10 pF probe on a 1 kΩ node adds a 16 MHz pole. On a crystal, a compensation node, or a high-impedance feedback divider, the circuit with the probe attached is a different circuit.

The probe is part of the circuit

Figure 1. Equivalent circuit of a probe on a node. The probe presents a resistance and a capacitance in parallel with the node, and the tip and ground lead add inductance in series with the measurement.

Every probe presents an input resistance $R_{IN}$ , an input capacitance $C_{IN}$ , and a lead inductance $L_{LEAD}$ to the node it touches. The resistance loads DC and low frequencies. The capacitance loads high frequencies and forms a low-pass filter with the source impedance of the node:

f_{-3dB} = \frac{1}{2 \pi R_S C_{IN}}

With $R_S = 1\ \text{k}\Omega$ and $C_{IN} = 10\ \text{pF}$ the corner is 16 MHz. With $R_S = 10\ \text{k}\Omega$ it is 1.6 MHz. Above the corner the probe attenuates the signal and shifts its phase, and the node itself behaves differently because the capacitance is now part of the circuit. The inductance forms a series resonance with $C_{IN}$ that is covered under the ground lead below.

Probe	$R_{IN}$	$C_{IN}$	Bandwidth	Maximum input	Use
Passive 1:1	1 MΩ	50 to 100 pF	10 to 25 MHz	300 V rms, derated with frequency	Low-level, low-frequency signals; ripple
Passive 10:1	10 MΩ	10 to 15 pF	200 to 500 MHz	300 V rms, derated with frequency	General purpose
Passive 100:1 high voltage	100 MΩ	3 to 6 pF	100 to 400 MHz	1 to 4 kV	Switch nodes above 300 V
Low-impedance (Z0) 10:1 or 20:1	500 Ω or 1 kΩ	0.5 to 1 pF	1 to 6 GHz	10 to 20 V	Fast digital, terminated lines, power rails
Active single-ended	100 kΩ to 1 MΩ	0.5 to 1 pF	1 to 8 GHz	±8 V signal, ±40 V offset typical	Fast signals on high-impedance nodes
Active differential	100 kΩ to 1 MΩ per side	0.3 to 1 pF	1 to 8 GHz	±5 V differential, low common mode	Differential signaling, small signals on a moving reference
High-voltage differential	4 to 10 MΩ per side	2 to 5 pF	50 to 200 MHz	1 to 7 kV differential and common mode	Gate drive, line-referenced nodes
Optically isolated	1 to 10 MΩ	1 to 2 pF	200 MHz to 1 GHz	60 kV common mode	High-side gate drive at fast edges
Current, Hall plus transformer	series insertion 10 to 50 mΩ and a few nH	-	20 to 150 MHz	30 to 150 A	DC to RF current
Rogowski coil	negligible insertion	-	30 MHz, no DC	kA	Large or inaccessible conductors

Voltage ratings on passive probes are quoted at DC or line frequency and derate with frequency, usually beginning between 100 kHz and 1 MHz and falling to a few tens of volts above 100 MHz. A 300 V probe on a 200 V, 100 kHz switch node is often outside its rating at the harmonics. Check the derating curve.

Passive probes

The 10:1 probe

Figure 2. Compensated 10:1 passive probe. The tip resistor and tip capacitor form one arm of the divider; the cable and scope capacitance with the 1 MΩ input form the other. The trimmer sets the capacitive ratio equal to the resistive ratio.

The 10:1 probe is a compensated divider. A 9 MΩ resistor in the tip and the scope's 1 MΩ input divide DC by ten. The cable and scope input contribute 60 to 100 pF, which by itself would roll the response off below 10 kHz, so a capacitor across the tip resistor provides a parallel capacitive divider. The two dividers match when

R_{TIP}\, C_{TIP} = R_{IN}\,(C_{CABLE} + C_{IN} + C_{COMP})

and the ratio is then 10:1 at every frequency. The trimmer $C_{COMP}$ is what the compensation adjustment sets. The node sees 10 MΩ in parallel with $C_{TIP}$ in series with the cable capacitance, which totals 10 to 15 pF, ten times better than the 1:1 probe and the reason the 10:1 probe is the default.

The cable is a lossy line, not a coax. A resistive center conductor damps the reflections that would otherwise ring on a 1.5 m unterminated cable. It also makes the cable capacitance part of the divider, which is why a probe is compensated for one scope input and why the probe's ratio drifts if the cable is changed.

Figure 3. Compensation. The calibrator square wave shows rounded corners when the tip capacitance is too small, a flat top when correct, and overshoot when too large.

Compensation is done on the scope's calibrator output, a 1 kHz square wave of about 1 to 3 V, with the probe on the channel it will be used with. Adjust the trimmer until the top of the square wave is flat. The adjustment matches the probe to that channel's input capacitance; a probe moved to another channel or another scope needs it again. An uncompensated probe reads correctly at DC and incorrectly at every frequency above the divider's crossover of a few kilohertz, so a 10 % compensation error is a 10 % amplitude error on a 10 kHz signal. Nothing on the display warns of it.

Some probes have a second, high-frequency adjustment in the compensation box. It is set with a fast-edge source and matched to the scope's input, and it is not touched routinely.

The 1:1 probe

A 1:1 probe is a length of cable with a tip. The node sees 1 MΩ in parallel with the full cable and scope capacitance, 50 to 100 pF, and the bandwidth is limited to 10 to 25 MHz by that capacitance and the cable loss. Its single advantage is the absence of attenuation: the scope's input noise is not multiplied by ten, so at 1 mV/div the noise floor is 100 to 300 µV rms instead of 1 to 3 mV rms at the tip. Switchable 1:1 / 10:1 probes have a 1:1 position with the 10:1 bandwidth reduced to the 1:1 figure; the switch position must match the channel setting.

	10:1	1:1
Loading	10 MΩ, 10 to 15 pF	1 MΩ, 50 to 100 pF
Bandwidth	200 to 500 MHz	10 to 25 MHz
Noise at tip	10 × scope noise	1 × scope noise
Maximum sensitivity at tip	10 mV/div for a 1 mV/div scope	1 mV/div
Use	Everything above a few MHz, any node above 1 kΩ	Ripple, small low-frequency signals, audio, sensor outputs

Low-impedance probes

Figure 4. Low-impedance 10:1 probe. A 450 Ω tip resistor drives a 50 Ω coax terminated in the scope's 50 Ω input. The tip capacitance is that of the resistor alone.

A resistor of 450 Ω or 950 Ω at the tip, a 50 Ω coax, and the scope's 50 Ω input make a 10:1 or 20:1 divider with no capacitive arm to compensate. The tip capacitance is that of a single surface-mount resistor, under 1 pF, and the bandwidth is set by the coax and the scope, typically several gigahertz. The cost is 500 Ω or 1 kΩ of resistive loading at all frequencies, which is acceptable on logic outputs, terminated transmission lines, power rails, and any node with source impedance below about 50 Ω, and unacceptable on high-impedance analog nodes. The maximum input is 10 to 20 V, set by the scope's 50 Ω input rating of 5 V rms. A Z0 probe can be made on the bench from a chip resistor soldered to the end of a semi-rigid coax, and this is the standard method for looking at a switch node or a fast logic edge without an active probe.

Active and differential probes

An active probe has an amplifier in the head, within a centimeter or two of the tip, so the cable capacitance is behind the amplifier and does not load the node. Input capacitance is 0.3 to 1 pF, input resistance is 100 kΩ to 1 MΩ, and bandwidth extends to several gigahertz. The limitations are the amplifier's: the linear input range is a few volts, offset range is a few tens of volts, the head is damaged by static and by overvoltage, and the probe needs power from the scope. An active probe on a 3.3 V logic node is the correct instrument; on a 48 V bus it is a destroyed probe.

Differential probes amplify the difference between two tips and reject what they have in common. Three figures describe one: the differential input range, the common-mode range, and the common-mode rejection ratio (CMRR) as a function of frequency. CMRR is quoted at DC or 60 Hz and falls with frequency, typically 80 dB at 60 Hz, 50 to 60 dB at 1 MHz, and 20 to 30 dB at 100 MHz. The error from imperfect rejection is the common-mode signal divided by the CMRR at the frequency of that signal, and a switch node has its energy at the edge rate, not the switching frequency.

Figure 5. Differential probe on a high-side gate. Both inputs ride on the switch node; the probe reports the gate-to-source voltage only as well as its CMRR at the edge rate allows.

High-voltage differential probes are the tool for gate drive and for any node referenced to a switching or line-connected point. They are resistive dividers of 50:1 or 500:1 into an amplifier with 1 to 7 kV common-mode range and 50 to 200 MHz bandwidth. Their CMRR at 100 MHz of about 30 dB means a 400 V switch node with a 10 ns edge produces a common-mode error on the order of 10 V during the edge, which is comparable to the gate voltage being measured. The measured $V_{GS}$ during the transition is therefore not trustworthy with a high-voltage differential probe; the plateau and the static levels are.

Optically isolated probes put the amplifier on the tip and send the signal to the scope over fiber. Common-mode range is tens of kilovolts, CMRR is above 100 dB at DC and stays above 80 dB at 100 MHz, and bandwidth reaches 1 GHz. They are the only instrument that shows a high-side gate waveform accurately during a fast edge, and they are priced accordingly.

Two channels and channel math is not a differential probe. A minus B on two single-ended probes has a CMRR set by the gain and delay matching of two separate channels and two separate probes, typically 30 to 40 dB at low frequency and near zero at high frequency, and both ground clips still go to earth. It is usable for a low-side sense resistor at low frequency and for nothing that moves.

The ground lead

Figure 6. Ground return. Left: the standard alligator lead forms a loop of tens of square centimeters with 100 to 200 nH of inductance. Right: a spring tip on the probe barrel closes the return within a centimeter of the tip.

The probe measures the voltage between its tip and its ground contact. The path from ground contact back to the barrel is a wire, and a wire has about 10 nH per centimeter, so the 15 cm alligator lead supplied with the probe adds 100 to 200 nH in series with the measurement. Two things follow.

Figure 7. Series resonance formed by the lead inductance and the probe capacitance. The source resistance of the node sets the damping.

Resonance. The lead inductance and the probe's 10 pF form a series LC circuit with

f_{ring} = \frac{1}{2 \pi \sqrt{L_{LEAD} C_{IN}}}, \qquad Q = \frac{\sqrt{L_{LEAD} / C_{IN}}}{R_S}

With 150 nH and 10 pF the resonance is 130 MHz and, on a 10 Ω source, the Q is 12. Any edge faster than a few nanoseconds excites it, and the display shows a 130 MHz ringing with tens of percent overshoot that does not exist on the board.

Figure 8. A 2 ns edge measured with a 15 cm ground clip and with a spring ground tip. The ringing on the upper trace is the probe resonance.

Pickup. The loop formed by the tip, the ground lead, and the board is an antenna. A switching converter has loops carrying amperes with nanosecond edges, and the magnetic field from those loops induces a voltage in the probe loop proportional to the loop area and $dI/dt$ . This appears on the trace as spikes at the switching edges regardless of what the tip is touching. The test for it is to connect the tip to the same point as the ground clip. Anything still on the screen is pickup, not signal.

The remedies, in order of effectiveness:

Spring ground tip. A short spring or wire from the barrel's ground ring to a ground point within a centimeter of the node. Inductance of a few nanohenries, resonance above 1 GHz, loop area near zero. Every probe ships with one.
Probe tip adapters soldered to the board: a pin for the tip and a ring for the barrel, or a small coax socket. Repeatable and hands-free, and the right approach for a node that will be measured repeatedly during development.
Coax pigtail soldered across the two points with 50 Ω termination at the scope. The best ground of all, with the loading of a 50 Ω or Z0 probe.
Short, twisted ground lead when the spring will not reach. Twisting reduces the loop area; length still sets the inductance.

The ground should return to the reference of the signal being measured, at the point where that signal is defined. For an output capacitor that is its negative terminal. For a gate drive it is the source pin. For a logic signal it is the nearest ground pin of the receiving device. Ground at the far end of the board and the measurement includes the ground drop between the two points.

Bandwidth, rise time, and sample rate

Analog bandwidth

Figure 9. Frequency response of a Gaussian scope front end. Amplitude is 3 dB low at the rated bandwidth and already 8 percent low at half of it.

The bandwidth on the front panel is the frequency at which a sine wave is displayed 3 dB, or 29 %, low. The rolloff below that is gradual. A scope with a Gaussian response reads 1 % low at one fifth of its bandwidth and 8 % low at half. Scopes above about 1 GHz use a flatter response with a sharper cutoff and smaller error below the corner, at the cost of overshoot on fast edges.

For sine waves, the rule of a bandwidth five times the highest frequency of interest keeps the amplitude error below about 2 %. For edges, the relevant quantity is rise time. A Gaussian system has

t_r \approx \frac{0.35}{BW}

and the rise times of the scope, the probe, and the signal add in quadrature:

t_{r,measured} = \sqrt{t_{r,signal}^2 + t_{r,scope}^2 + t_{r,probe}^2}

A 500 MHz scope with a 500 MHz probe has a system rise time of about 1 ns. It reports a 1 ns edge as 1.4 ns, a 3 ns edge as 3.2 ns, and a 10 ns edge as 10.05 ns. To measure an edge within 5 % the system rise time must be under a third of it. A GaN half bridge with a 2 ns edge needs a system rise time below 0.7 ns, which is a 1 GHz scope and probe.

Bandwidth is also a noise setting. The scope's input noise is broadband, and its rms value scales with the square root of the bandwidth. A 500 MHz front end at 1 mV/div might show 200 µV rms; the 20 MHz limit reduces that by a factor of five to 40 µV rms.

Sample rate

The sample rate is a property of the digitizer and has nothing to do with the analog bandwidth. A sampled record represents frequencies below half the sample rate. Frequencies above that fold back into the band as aliases, and nothing in the record distinguishes an alias from a real signal.

Figure 10. Aliasing. A signal at 0.9 times the sample rate produces samples identical to a signal at 0.1 times the sample rate. The scope draws the alias.

Two conditions must hold. The sample rate must exceed twice the highest frequency the front end passes, and the display interpolation must have enough points per cycle. With sin(x)/x interpolation, 2.5 samples per period of the bandwidth are sufficient, which is why a 1 GHz scope has a 2.5 GS/s digitizer. With linear interpolation, 10 samples per period are needed for the display to resemble the signal.

The trap is the timebase. The scope's memory holds a fixed number of points, and the sample rate is the memory depth divided by the record length:

f_s = \frac{\text{memory depth}}{10 \times \text{time/div}}

A scope with 10 Mpoints at 1 GS/s holds 10 ms. At 100 ms/div the record is 1 s and the sample rate drops to 10 MS/s while the analog bandwidth stays at 500 MHz. Everything between 5 MHz and 500 MHz aliases. A switching converter viewed at 100 ms/div to look at a load transient shows a switching-frequency alias as a slow beat that looks like output oscillation. The indications are a displayed sample rate that has dropped, and a waveform whose apparent frequency changes when the timebase changes.

Remedies:

Read the sample rate on the display at the timebase in use, and check that it exceeds 2.5 times the front-end bandwidth or the bandwidth limit is on.
Peak detect acquisition keeps the maximum and minimum within each sample interval. The switching ripple appears as a band of the correct height at any timebase. Amplitude is preserved, waveform shape is not.
High-resolution acquisition averages the samples within each interval, which filters the alias and adds resolution. Ripple disappears from the display, which is correct if the intent is to see the average.
Deeper memory or a zoomed-in second timebase for the detail.

The bandwidth limit

The 20 MHz limit (some scopes offer 200 MHz and 250 MHz as well) is a low-pass filter ahead of the ADC. It is the right setting when:

The measurement is defined at 20 MHz. Regulator ripple and noise specifications are written that way, and a measurement at 500 MHz reads higher because it includes the switch-edge spikes and the probe pickup.
The signal is slow and small. Sensor outputs, audio, control-loop signals, and reference voltages carry nothing above 20 MHz, and the limit removes five times the rms noise.
A wideband aliasing problem must be cut off at a slow timebase, as above.

It is the wrong setting when anything of interest has content above 20 MHz: switch-node edges and overshoot, gate drive, ringing, logic edges, clock signals, and any timing measurement below about 50 ns. The filter turns a 5 ns edge into 17 ns and removes overshoot entirely. The limit stays off by default and is turned on for a specific measurement, and the state is recorded with the result because a ripple figure is meaningless without its bandwidth.

Vertical resolution

Figure 11. Vertical resolution. The ADC quantizes the full screen. A signal using two divisions is digitized with a quarter of the codes available to one using seven.

The ADC digitizes the full screen. An 8-bit converter has 256 codes across eight vertical divisions, 32 codes per division, and one code is the volts-per-division setting divided by 32. That is the resolution regardless of where the signal is. A signal that occupies two divisions is digitized with 64 codes, and its smallest visible step is 1.6 % of its amplitude. The same signal filling seven divisions has 224 codes and a step of 0.45 %.

ADC	Codes	Codes per division (8 div)	Step at 100 mV/div
8 bit	256	32	3.1 mV
10 bit	1024	128	0.78 mV
12 bit	4096	512	0.20 mV

The stated resolution is a ceiling. Front-end noise and distortion reduce the effective number of bits, and an 8-bit scope at full bandwidth typically delivers 6 to 7 effective bits; a 12-bit scope delivers 9 to 11 depending on bandwidth and sensitivity.

Rules for using the bits:

Fill the screen. Set volts per division so the signal spans six to seven of eight divisions. This is the single largest improvement available and costs nothing.
Use offset, not position, for a signal riding on a DC level. Offset subtracts a voltage ahead of the amplifier so a small ripple on a 12 V rail can be displayed at 10 mV/div. Position moves the trace after digitizing and does not help. Offset range is limited and depends on the volts-per-division setting; when it is not enough, AC coupling is the alternative.
Do not clip. A signal beyond the screen drives the input amplifier out of its linear range, and its recovery corrupts the visible part of the trace for a time after the overload. Measurements on a clipped acquisition are wrong even in the parts that appear on screen.
Zoom does not add resolution. Expanding a stored acquisition shows the codes it was digitized with. Resolution is set at acquisition.
High-resolution mode averages consecutive samples within one acquisition. Averaging $N$ samples adds $\tfrac{1}{2}\log_2 N$ bits, so each halving of the effective sample rate adds half a bit. The cost is bandwidth, and it works on single-shot signals.
Averaging mode averages successive acquisitions and requires a repetitive signal with a stable trigger. It reduces random noise by $\sqrt{N}$ and removes anything not synchronous with the trigger, including real jitter.
Probe attenuation costs resolution. A 10:1 probe at 10 mV/div at the tip is the scope at 1 mV/div, with the scope's noise at that setting multiplied by ten. A 1:1 probe or coax gives the same tip sensitivity at 10 mV/div on the scope.

Coupling, termination, and the input path

Figure 12. The vertical channel. Coupling, termination, and the bandwidth limit are ahead of the ADC, so a wrong setting is part of the record.

DC coupling is the default and shows the signal as it is. AC coupling inserts a series capacitor that with the 1 MΩ input forms a high-pass filter with a corner near 10 Hz. It removes the DC component so that a small signal on a large rail can be displayed at high sensitivity without offset range. Its side effects: a square wave below a few hundred hertz tilts as the coupling capacitor charges during each half cycle; the trace takes a fraction of a second to settle after a DC step, such as a load step or turning the supply on; and on most scopes AC coupling is unavailable with the 50 Ω input. For a load transient, where the response is the DC level moving, DC coupling with offset is the correct choice and AC coupling distorts the result.

Termination. The 1 MΩ input is for passive probes and for anything driven through a cable at low frequency. The 50 Ω input terminates a coax, a Z0 probe, or an active probe's output in its characteristic impedance and is required for those; a coax into 1 MΩ reflects and rings. The 50 Ω input is rated for about 5 V rms and is destroyed by more. Some scopes lock out 50 Ω above a few volts per division for that reason.

Attenuation setting. The channel must know the probe ratio so that the displayed volts are correct. Probes with a sense pin on the BNC set it automatically; probes without one, and every homemade coax or Z0 probe, require it to be set by hand, and the setting persists after the probe is removed.

Probe deskew. Different probes have different propagation delays, 5 to 15 ns for a passive probe, more for a current probe. When two channels are compared in time, or multiplied to compute power, the channels must be deskewed with a common fast-edge source. A 5 ns skew between a voltage probe and a current probe is a large error in a switching-loss measurement where the transition lasts 20 ns.

Measuring ripple and noise

Figure 13. Output ripple measurement. The tip on the capacitor positive pad, the ground spring on the negative pad, and nothing else in the loop.

Ripple is a small signal on a large DC level, at the switching frequency and its harmonics, accompanied by spikes at the switching edges. The measurement is defined by three settings and one mechanical detail.

Location. Across the output capacitor, or across the point the specification names. The tip on the positive pad, the ground spring on the negative pad. Nothing else in the loop.
Bandwidth. The figure quoted in a datasheet is at 20 MHz unless stated otherwise. Turn the limit on for that measurement and record it. Measure again at full bandwidth to see the edge spikes; the difference between the two is the high-frequency content that the ground lead and the layout will determine on the final product.
Coupling and sensitivity. AC coupling, or DC coupling with offset if the scope has enough, and volts per division set so the ripple fills the screen.
Probe. A 1:1 probe, a 1:1 coax with a DC block into 50 Ω, or a dedicated power-rail probe. A 10:1 probe multiplies the scope's noise floor by ten at the tip; at 10 mV/div that noise is comparable to the ripple of a good regulator.

The noise floor is checked by touching the tip to the ground spring's contact point. What the screen shows then is the floor of the scope, the probe, and the pickup, and the ripple measurement is only valid to the extent that it is well above this.

A power-rail probe is an active probe with a 50 kΩ input, ±60 V or so of built-in offset, and 1 to 4 GHz of bandwidth at 1:1 attenuation. It exists because the alternatives, AC coupling with limited offset or a 1:1 passive probe with 20 MHz of bandwidth, cannot show a 10 mV disturbance on a 1.8 V rail at the bandwidth that a modern digital load has.

Switch node and gate drive

Switch node. The quantities of interest are overshoot, ringing frequency, and edge rate, all of which are high-frequency. Full bandwidth, spring ground tip on the source of the low-side switch or on the nearest ground pad, and a probe rated for the voltage at the harmonics present. Below 100 V a 10:1 passive probe on a spring tip is adequate; above that a 100:1 high-voltage passive probe with the same care about grounding. The measured overshoot includes the probe's own resonance unless the spring tip is used, and 20 to 30 % of apparent overshoot on a switch node is commonly the ground lead.

Low-side gate. Tip on the gate pin, spring ground on the source pin of the same package. The gate voltage referenced anywhere else includes the source inductance drop, which during a fast transition is volts, and the waveform shows a spurious dip or a Miller plateau that is not there.

High-side gate. The source is the switch node. The measurement must be differential across the gate and source, and both inputs move through the full bus voltage at the edge rate. The candidates, from least to most accurate at the edge:

Method	Valid for	Not valid for
Two single-ended probes, A minus B	Nothing on a switching node; the ground clips short the switch node to earth	Everything
High-voltage differential probe, 100 MHz	Static levels, plateau voltage, timing of the transition to within tens of ns	$V_{GS}$ during the edge, ringing on the gate
Optically isolated probe, 1 GHz	The waveform during the edge	Cost-limited
Isolated scope channels (battery-powered, or per-channel isolation)	Static levels and slow edges; CMRR at high frequency is poor	Fast edges

The evaluation of a differential measurement is the same as for ripple: connect both tips to the switch node and look at what the probe reports. That is the common-mode error at the actual edge rate, and the gate measurement is valid only where the signal is well above it.

Current measurement

Method	DC	Bandwidth	Insertion	Notes
Hall plus transformer clamp probe	Yes	20 to 150 MHz	10 to 50 mΩ, a few nH	Needs a loop of wire through the jaw; degauss and zero before use; DC drift with temperature; core saturates on overload
Current transformer	No	1 kHz to 200 MHz typical	Small, a burden resistor reflected by $1/N^2$	Inserted in the loop permanently; no DC; low-end corner set by magnetizing inductance and burden
Rogowski coil	No	Hz to 30 MHz	Negligible	Flexible, fits around bus bars and packages; integrator drift; low sensitivity
Sense resistor with voltage probe	Yes	Set by the resistor's inductance and the probe	The resistor	Kelvin connections; 1 nH in a 10 mΩ resistor is a corner at 1.6 MHz, so use a low-inductance part or compensate; the measurement is single-ended only if one end is ground
Shunt with differential probe	Yes	Set by the probe	The resistor	High-side sensing; CMRR limits apply at the switching edges

A current probe requires a loop of wire to clamp around. The loop adds inductance to the circuit, typically 10 to 30 nH for a 3 cm loop, which is often more than the parasitic inductance of the layout being measured and changes the waveform being observed, particularly the current overshoot at a switching edge. The loop belongs in a slow path, such as the inductor lead, not in the switching loop. Current probes also have a delay of 10 to 20 ns that must be deskewed against the voltage channel before computing power.

Floating measurements and safety

Figure 14. The floating-scope hazard. The probe ground clip is connected to protective earth through the scope chassis. Clipping it to the primary return of an offline supply puts rectified line voltage across that path.

A bench scope's chassis, its BNC shells, and every probe ground clip are bonded to protective earth. Connecting a ground clip to any node that is not at earth potential shorts that node to earth through the probe lead and the scope. On a low-voltage board the result is a blown trace or a damaged probe. On the primary side of an offline converter the node is rectified line, 160 to 400 V from earth, and the result is a destroyed scope, a destroyed board, and a shock hazard.

The rules:

The scope stays earthed. Removing the earth pin, or running the scope on an isolation transformer, makes the entire chassis and every BNC live at the potential of whatever the ground clip touches. It is done, and it is the most common cause of scope-related injury.
The isolation transformer goes on the DUT, if one is used at all. The DUT's primary return can then be tied to earth through the probe ground with no fault current. This is a measurement convenience only; the DUT's own reference is still not earth, and every other connection to it (a computer, another instrument) must be considered.
Differential and isolated probes are the correct instrument for any line-referenced or switching-referenced node. Their common-mode voltage rating must exceed the node's potential from earth, including transients.
Category ratings. CAT II, III, and IV describe the transient overvoltage the probe withstands at its rated voltage, from a wall outlet (II) to the distribution panel (III) to the service entrance (IV). A 300 V CAT II passive probe is not a 300 V CAT III probe. Mains-connected work uses CAT III probes with the finger guard in place.
Passive probe voltage derating applies. A 1 kV probe is rated at DC and line frequency, and at 1 MHz its rating may be 100 V. On a switch node the harmonics are at megahertz.

Procedure before trusting a measurement

The probe is compensated on the channel in use, and the channel's attenuation setting matches the probe.
The scope's bandwidth is at least five times the highest frequency of interest, or the system rise time is under a third of the edge being measured.
The bandwidth limit is off, unless the measurement is defined at 20 MHz, and its state is recorded with the result.
The sample rate shown on the display is at least 2.5 times the front-end bandwidth at the timebase in use.
The ground return is a spring tip or an adapter at the node, not the alligator lead, and it returns to the reference of the signal being measured.
The signal fills six to seven of eight divisions without clipping.
The tip is touched to the ground contact and the screen shows only the noise floor. Anything else is pickup, and the measurement is valid only well above it.
For a differential measurement, both tips on the common-mode node shows only the common-mode error at the actual edge rate.
The circuit behaves the same with the probe attached and removed. If it does not, the probe capacitance or its 500 Ω is part of the circuit, and a lower-capacitance probe or a buffered test point is needed.
The probe's voltage rating at the frequencies present, and its category rating, exceed the node.