Optically isolated probes and high-CMRR measurement

Measuring 15 V of gate drive on a node that swings 600 V in ten nanoseconds. Why an ordinary differential probe cannot do it, how power-over-fibre can — and the part the datasheets leave out: the attenuator you fit decides your bandwidth, your voltage range and your common-mode rejection, all at once.

16 min read

Every hard measurement in a switching converter is the same measurement: a small differential signal sitting on a large, fast-moving common-mode voltage. The canonical case is the high-side gate of a half-bridge. You want V_GS, which is about 15 V. Its reference terminal is the switch node, which swings the entire DC link in a few nanoseconds. The signal you want and the signal you do not want share the same two probe tips.

The high-side gate measurement. The 15 V you want is measured between two points that are together moving 600 V in about ten nanoseconds.

Common-mode rejection, converted into volts

Common-mode rejection ratio is the number that says how much of that unwanted swing leaks into the result. It is quoted in decibels, which makes it easy to nod at and hard to feel. Convert it into volts and it stops being abstract.

A classical high-voltage differential probe has roughly 25 dB of common-mode rejection at 100 MHz — a ratio of about 18:1. Put the half-bridge above in front of it:

Quantity	Value	Where it comes from
Common-mode voltage at the switch node	600 V	the DC link
Differential signal you want	15 V	the gate drive
Probe CMRR at 100 MHz	25 dB ≈ 18:1	Wolfspeed, Dynamic Characterization and Measurement Methods
Error the probe contributes	≈ 33 V	600 ÷ 18

The error is more than twice the signal — and nothing about the trace tells you which part is which. This is why high-side gate waveforms measured with an ordinary differential probe show ringing that the circuit is not actually producing, and why the switching losses computed from them are wrong.

What optical isolation actually removes

A differential amplifier subtracts two inputs that are both referenced, eventually, to the oscilloscope’s ground and the mains earth behind it. That shared reference is a conductive path, and common-mode current flows in it. An optically isolated probe does not improve the subtraction — it removes the path.

The difference is not a better amplifier. It is the absence of a conductor: the front end runs from its own battery and the only link to the oscilloscope is light.

Front end (E/O transmitter) — takes the attenuator, runs from a replaceable lithium battery for about 8 hours, and sits on the circuit at whatever potential the circuit is at
Two metres of optical fibre — the only thing joining the two halves, and the entire reason the isolation exists
Rear end (O/E receiver) — plugs into an oscilloscope BNC, takes USB 5 V, and presents a 50 Ω terminated ±0.5 V output

The isolation voltage that results is ±60 kV on the ODP6000B series. That figure is not the signal you measure — it is the potential the probe tip may float at relative to the oscilloscope while you measure.

⚠ The three specifications that trade against each other

Here is what the factory literature does not put in front of you. An optically isolated probe has no fixed measurement range, no fixed common-mode rejection and no usable-bandwidth figure of its own. All three are set by the attenuator you screw onto the tip, and they do not move in the same direction.

Start with the range, because it is the simplest. The probe amplifier itself accepts ±0.5 V. Every measurement range in the ODP6000B family is that figure multiplied by the attenuation ratio — which is why the spec sheet’s headline “±25 V” is really the 50X number, not a property of the probe.

Now the part that decides whether the measurement is any good. Siglent publish the common-mode rejection as a graph with no numbers attached, which makes it impossible to plan around. Read off that chart and tabulated, it looks like this:

Common-mode rejection against frequency, by attenuator. Every tip starts near 180 dB at DC; the high-voltage tips are the ones that fall off a cliff.

Attenuator	DC	1 MHz	10 MHz	100 MHz	1 GHz
5X	≈180	≈168	≈150	≈130	≈90	option
50X	≈180	≈150	≈125	≈100	≈90	supplied
200X	≈180	≈135	≈115	≈85	≈80	option
1000X	≈180	≈100	≈85	≈55	≈30	supplied
2000X	≈180	≈90	≈80	≈50	≈20	supplied
5000X	≈180	≈85	≈75	≈45	≈15	supplied
10000X	≈180	≈80	≈70	≈40	≈10	option

Values in dB, read from the CMRR curves in the ODP6000B instruction manual and rounded to the nearest 5 dB. Treat them as indicative rather than guaranteed — but they are the only numeric form of this data that exists.

Read the 1 GHz column. With the 50X tip you still have about 90 dB. With the 5000X tip — the one you would reach for on a kilovolt rail — you have about 15 dB, a rejection of less than six to one. Put 600 V of common mode on that and roughly 100 V of it lands in your result. The gigahertz and the kilovolts are available from the same probe, but not from the same measurement.

The voltage-versus-frequency limit

The second constraint is a hard one, published in the manual and absent from almost every listing of these probes: the maximum voltage you may apply falls as the test frequency rises. Plotted as a boundary, the shape of the problem is obvious.

The published maximum-test-voltage limit. Everything above and to the right of the line is outside specification — and the line never reaches 500 MHz, let alone 1 GHz.

Maximum test voltage	Maximum test frequency	Attenuator
±5000 Vpk	700 kHz	10000X
±2500 Vpk	800 kHz	5000X
±1000 Vpk	2 MHz	2000X
±500 Vpk	3 MHz	1000X
±250 Vpk	20 MHz	500X
±100 Vpk	50 MHz	200X
±50 Vpk	100 MHz	100X
±25 Vpk	200 MHz	50X

The highest row in that table is 200 MHz, and it is the 50X tip at ±25 Vpk. At full voltage you have 700 kHz. A 500 MHz or 1 GHz optically isolated probe is a small-signal instrument that also happens to survive being floated at 60 kV — not a high-voltage instrument with a gigahertz of bandwidth.

Choosing the attenuator: a procedure, not a parts list

Because all three specifications hang off this one choice, it is worth doing deliberately rather than fitting whatever is on the bench.

1. Find the real peak differential voltage, including overshoot — not the nominal gate voltage. A 15 V gate drive that rings to 22 V needs headroom for 22 V.
2. Take the lowest ratio whose range covers it. 22 V fits the 50X tip (±25 Vpk) with nothing to spare, so on a noisy board go to 100X.
3. Check that ratio against the derating boundary at your highest frequency of interest. If the ratio you picked is capped below the bandwidth you need, the measurement cannot be made as specified — reduce the signal, not the expectation.
4. Check the CMRR at that ratio and frequency against your common-mode voltage. Divide: common mode ÷ 10^(CMRR/20) is the error in volts. If it is not small next to your signal, the measurement will not mean anything.
5. Never fit a higher ratio “for safety”. Every step up costs common-mode rejection, bandwidth headroom and vertical resolution together.

Worked example. High-side V_GS peaking at 20 V, 600 V rail, edges of about 2 ns so interest out to ~150 MHz. Step 2 gives the 50X tip (±25 Vpk). Step 3: the 50X row is good to 200 MHz — fine. Step 4: about 95 dB at 150 MHz, so 600 V ÷ 10^4.75 ≈ 11 mV of error against a 20 V signal. That measurement is worth trusting.

Which probe: 500 MHz or 1 GHz?

A probe does not measure a rise time; it measures the combination of its own and the signal’s. To a good first approximation the two add in quadrature, so the measured value is √(t_signal² + t_probe²). Put the two published rise times — 0.7 ns for the ODP6050B, 0.45 ns for the ODP6100B — through that and the choice makes itself.

Your signal’s real rise time	ODP6050B shows	error	ODP6100B shows	error
5 ns	5.05 ns	+1 %	5.02 ns	+0.4 %
2 ns	2.12 ns	+6 %	2.05 ns	+2.5 %
1 ns	1.22 ns	+22 %	1.10 ns	+10 %
0.5 ns	0.86 ns	+72 %	0.67 ns	+35 %

Edges slower than about 2 ns — the ODP6050B is within ~6 % and the extra €4,030 buys very little. Spend it on attenuators.
Edges around 1 ns, typical of SiC gate drive — the 500 MHz probe reads 22 % slow. This is where the ODP6100B earns itself.
Edges near 0.5 ns — ⚠ neither probe is accurate. The ODP6100B is still 35 % out. Buy it for the isolation, not for sub-nanosecond fidelity.

Your oscilloscope’s own rise time adds into the same sum, so the real error is a little larger again than the table shows.

Setting it up so the answer is right

Set the channel to 50 Ω. The receiver is a 50 Ω terminated source. The manual requires it whenever the measured rise time is under 3.5 ns — which, on a probe bought for this work, is always.
Set the channel’s attenuation ratio to match the tip fitted. Get this wrong and every voltage on screen is wrong by that factor, silently.
Auto-zero after the probe reaches temperature, and again after any attenuator change. It takes about 20 seconds and does not require disconnecting from the circuit.
Suspend the transmitter on its bracket, away from the high-voltage pulse circuitry. It is a floating box next to a fast dv/dt source; where you put it changes what it picks up.
Keep the fibre bend radius above 10 cm, and inspect it before each use. The fibre is the isolation.
Switch the circuit off before removing the probe. The front end is connected directly to high voltage on the device under test.

Practical trap: with an attenuator fitted, it mechanically blocks the battery door. Remove the tip before trying to change batteries, rather than forcing the cover.

What a good high-side turn-on looks like

Once the measurement is trustworthy, the gate waveform is readable in three parts. First the C_GS charging ramp, as the driver charges the gate-source capacitance. Then the Miller plateau — a flat region where the drain voltage is moving and the driver current is going into C_GD instead of raising V_GS; its length is a direct measure of how long the device spends in transition, and therefore of switching loss. Finally the gate charges to its full drive voltage.

The reason this note exists is that the Miller plateau is exactly where a probe with insufficient common-mode rejection lies to you: the plateau is flat and low-amplitude while the drain is slewing fastest, so it coincides with the largest common-mode transient of the whole cycle. Ringing painted onto the plateau by a poor probe is routinely mistaken for gate-loop oscillation.

When not to buy one

Optical isolation is expensive, and it is not always what the measurement needs. This is the honest boundary:

	DPB6150D differential probe	ODP6050B / ODP6100B optical
Bandwidth	400 MHz	500 MHz / 1 GHz
CMRR	> 80 dB at DC, falling steeply with frequency	≈ 180 dB at DC; see the curves above
Differential range	±1500 Vpk	±2.5 V to ±5000 Vpk, by attenuator
Isolation voltage	—	±60 kV
Price ex VAT	€1,420	€4,500 / €8,530

Below about 400 MHz with moderate common mode — a differential probe measures the same signal for a fraction of the cost. Buy that.
Low-side gate drive, or anything ground-referenced — a 10:1 passive probe is fine. There is nothing to isolate.
High-side gate drive on fast SiC or GaN — this is the case optical isolation was built for, and where nothing cheaper gives an honest answer.

The short version: an optically isolated probe does not measure smaller signals than a differential probe — it measures them while the reference moves. Buy it for the common-mode problem, size the attenuator for the signal, and check the derating boundary before you promise anyone a bandwidth.